diff --git a/physiological_integration/cases/case1/case1_01.json b/physiological_integration/cases/case1/case1_01.json new file mode 100644 index 0000000000000000000000000000000000000000..28688f089e6245f0ec3f6e3a1fd14a37697b7b33 --- /dev/null +++ b/physiological_integration/cases/case1/case1_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Inspect the provided recording and determine whether the channel named \"Fp1\" and the channel named \"Status\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_02.json b/physiological_integration/cases/case1/case1_02.json new file mode 100644 index 0000000000000000000000000000000000000000..3a1a331922138bece71f5cdc9c534f02bd7c14b5 --- /dev/null +++ b/physiological_integration/cases/case1/case1_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Inspect the provided recording and determine whether the channel named \"GSR1\" and the channel named \"Resp\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_03.json b/physiological_integration/cases/case1/case1_03.json new file mode 100644 index 0000000000000000000000000000000000000000..bb62b2f109471e87c62c4f74019c96c1f6879d76 --- /dev/null +++ b/physiological_integration/cases/case1/case1_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Inspect the provided recording and determine whether the channel named \"Plet\" and the channel named \"Temp\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_04.json b/physiological_integration/cases/case1/case1_04.json new file mode 100644 index 0000000000000000000000000000000000000000..54075a9f3a9391ed92549bb9ad486d71696350b1 --- /dev/null +++ b/physiological_integration/cases/case1/case1_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Inspect the provided recording and determine whether the channel named \"GSR2\" and the channel named \"ECG\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_05.json b/physiological_integration/cases/case1/case1_05.json new file mode 100644 index 0000000000000000000000000000000000000000..ea804ccf296915d9e4a82501cc56a16cb101b8a5 --- /dev/null +++ b/physiological_integration/cases/case1/case1_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Inspect the provided recording and determine whether the channel named \"Resp\" and the channel named \"S01D01_760\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_06.json b/physiological_integration/cases/case1/case1_06.json new file mode 100644 index 0000000000000000000000000000000000000000..fab62fa60e97bae9a4daa66452004d765b7329bd --- /dev/null +++ b/physiological_integration/cases/case1/case1_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"FT7\" and the channel named \"eog_h\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_07.json b/physiological_integration/cases/case1/case1_07.json new file mode 100644 index 0000000000000000000000000000000000000000..2647e0077f35345ff5f75b30e8d56e8b298cf4b7 --- /dev/null +++ b/physiological_integration/cases/case1/case1_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"eog_v\" and the channel named \"perclos\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_08.json b/physiological_integration/cases/case1/case1_08.json new file mode 100644 index 0000000000000000000000000000000000000000..ba45fa56612f824087aca828138ab12b71768f52 --- /dev/null +++ b/physiological_integration/cases/case1/case1_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"EOG-H\" and the channel named \"GSR1\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_09.json b/physiological_integration/cases/case1/case1_09.json new file mode 100644 index 0000000000000000000000000000000000000000..6ede173a0a399b6591ce4173b76f46b754d36e98 --- /dev/null +++ b/physiological_integration/cases/case1/case1_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"POZ\" and the channel named \"Status\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_10.json b/physiological_integration/cases/case1/case1_10.json new file mode 100644 index 0000000000000000000000000000000000000000..777234699f76d261494871a2806a8acf913b5390 --- /dev/null +++ b/physiological_integration/cases/case1/case1_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"O2\" and the channel named \"eog_config\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_11.json b/physiological_integration/cases/case1/case1_11.json new file mode 100644 index 0000000000000000000000000000000000000000..5ad5aaf8b1ff1853a01aac28430e5ac4c87adfc3 --- /dev/null +++ b/physiological_integration/cases/case1/case1_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"SaO2\" and the channel named \"ECG\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_12.json b/physiological_integration/cases/case1/case1_12.json new file mode 100644 index 0000000000000000000000000000000000000000..eafd0a543c27312ec619d13ec9769e5481109dff --- /dev/null +++ b/physiological_integration/cases/case1/case1_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"EMG\" and the channel named \"EOG(L)\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_13.json b/physiological_integration/cases/case1/case1_13.json new file mode 100644 index 0000000000000000000000000000000000000000..079030f5b17aff9fc4149fedccdd3ba6eb7c9c2a --- /dev/null +++ b/physiological_integration/cases/case1/case1_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"THOR RES\" and the channel named \"OX stat\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_14.json b/physiological_integration/cases/case1/case1_14.json new file mode 100644 index 0000000000000000000000000000000000000000..838d30359a41174a9f53f611051cc99668133631 --- /dev/null +++ b/physiological_integration/cases/case1/case1_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"EEG(sec)\" and the channel named \"Status\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_15.json b/physiological_integration/cases/case1/case1_15.json new file mode 100644 index 0000000000000000000000000000000000000000..f06b57bb9e4adc73e59f85d75c2faee7a2900a13 --- /dev/null +++ b/physiological_integration/cases/case1/case1_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case01_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Inspect the provided recording and determine whether the channel named \"H.R.\" and the channel named \"AIRFLOW\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_16.json b/physiological_integration/cases/case1/case1_16.json new file mode 100644 index 0000000000000000000000000000000000000000..ea4b08630cc8077870f73eae7d38a58601af9e1c --- /dev/null +++ b/physiological_integration/cases/case1/case1_16.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"S01D01_760\" and the channel named \"HEOG\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_17.json b/physiological_integration/cases/case1/case1_17.json new file mode 100644 index 0000000000000000000000000000000000000000..8fedb4c823a66d28a139cf6e223ecfb498bb8b44 --- /dev/null +++ b/physiological_integration/cases/case1/case1_17.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"S01D01_850\" and the channel named \"VEOG\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_18.json b/physiological_integration/cases/case1/case1_18.json new file mode 100644 index 0000000000000000000000000000000000000000..6f86cebb9ee3c991b51481e96b42782593f61ff9 --- /dev/null +++ b/physiological_integration/cases/case1/case1_18.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"FP1\" and the channel named \"S16D16_850\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_19.json b/physiological_integration/cases/case1/case1_19.json new file mode 100644 index 0000000000000000000000000000000000000000..49835fa5c6832f98c09f5ec90e152992bf4279e0 --- /dev/null +++ b/physiological_integration/cases/case1/case1_19.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"S01D01_760\" and the channel named \"S 32\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_20.json b/physiological_integration/cases/case1/case1_20.json new file mode 100644 index 0000000000000000000000000000000000000000..fc878c6e740e2a1d33c78e6632ddba521a6673da --- /dev/null +++ b/physiological_integration/cases/case1/case1_20.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"HEOG\" and the channel named \"SaO2\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_21.json b/physiological_integration/cases/case1/case1_21.json new file mode 100644 index 0000000000000000000000000000000000000000..cd1c3615fbaeb1b7105466fc14ca6bc538d095f4 --- /dev/null +++ b/physiological_integration/cases/case1/case1_21.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"FP1\" and the channel named \"Gaze point X\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_22.json b/physiological_integration/cases/case1/case1_22.json new file mode 100644 index 0000000000000000000000000000000000000000..3af6c8f5a7c15e8f006604ae26e2a0188f106f69 --- /dev/null +++ b/physiological_integration/cases/case1/case1_22.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"Pupil diameter left\" and the channel named \"timestamps_sec\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_23.json b/physiological_integration/cases/case1/case1_23.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec03d73274f6f27d2ac4dabada990b8970ca9f8 --- /dev/null +++ b/physiological_integration/cases/case1/case1_23.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"HEO\" and the channel named \"Gaze point Y\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_24.json b/physiological_integration/cases/case1/case1_24.json new file mode 100644 index 0000000000000000000000000000000000000000..b91ee47fe947624133eb73a3fb37365cf3af67b8 --- /dev/null +++ b/physiological_integration/cases/case1/case1_24.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"ECG\" and the channel named \"validity_left\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case1/case1_25.json b/physiological_integration/cases/case1/case1_25.json new file mode 100644 index 0000000000000000000000000000000000000000..2eff97aead5dc095c98baa6cfbebf7ff6698b059 --- /dev/null +++ b/physiological_integration/cases/case1/case1_25.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case01_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "instruction": "Inspect the provided recording and determine whether the channel named \"Pupil diameter right\" and the channel named \"trigger\" are analyzable physiological signal channels. Treat event markers, status fields, labels, timestamps, and other non-physiological fields as absent. Return only channel_1_exists and channel_2_exists as JSON booleans; false means that the requested name is absent or is not an analyzable physiological signal." + }, + "eval_config": { + "parser_prompt": "## ROLE\nYou are a precise JSON extraction engine for neurophysiological data. Convert the agent report into one JSON object.\n\n## TASK\nExtract the final existence decision for the two requested channel names:\n- channel_1_exists: whether the first requested channel is an analyzable physiological signal channel.\n- channel_2_exists: whether the second requested channel is an analyzable physiological signal channel.\n\n## STRICT CONSTRAINTS (MANDATORY)\n1. Output only one valid JSON object; do not use Markdown fences or explanations.\n2. The object must contain exactly the keys `channel_1_exists` and `channel_2_exists`.\n3. Both values must be JSON booleans (`true` or `false`), not strings or numbers.\n4. Use `false` when a requested name is absent or identifies a status, event, label, timestamp, marker, or other non-physiological field.\n5. Do not output modality names, channel names, evidence, or any additional fields.\n6. If the report omits an existence decision, output `false` for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_exists\": true, \"channel_2_exists\": false}", + "metrics": [ + { + "metric_id": "channel_1_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_1_exists", + "weight": 50, + "params": { + "gt_value": true + } + }, + { + "metric_id": "channel_2_exists_accuracy", + "type": "categorical_check", + "target_key": "channel_2_exists", + "weight": 50, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_01.json b/physiological_integration/cases/case10/case10_01.json new file mode 100644 index 0000000000000000000000000000000000000000..cafce314253e1b320c2a9139b29fe887e1676ef1 --- /dev/null +++ b/physiological_integration/cases/case10/case10_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "From respiration channel `THOR RES`, analyze the half-open interval `[0, 120)` seconds at a sampling rate of `10 Hz`.\n\nApply the declared fourth-order `0.05–0.7 Hz` band-pass filter.\nStandardize the filtered signal, then detect peaks with a minimum distance of 20 samples and a minimum prominence of 0.5.\nFor each pair of consecutive peaks, convert the peak interval to breaths per minute as `60 / interval_seconds`. Set `feature_value` to the arithmetic mean of these interval-based respiration rates.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 22.834861, + "tolerance": 2.2834861 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_02.json b/physiological_integration/cases/case10/case10_02.json new file mode 100644 index 0000000000000000000000000000000000000000..9f6914c22498d0864e8cea14707ece2769b8c724 --- /dev/null +++ b/physiological_integration/cases/case10/case10_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "From respiration channel `ABDO RES`, analyze the half-open interval `[300, 420)` seconds at a sampling rate of `10 Hz`.\n\nApply the declared fourth-order `0.05–0.7 Hz` band-pass filter.\nStandardize the filtered signal, then detect peaks and preceding troughs with a minimum peak distance of 20 samples and a minimum prominence of 0.5.\nFor each detected peak, measure its amplitude relative to the preceding trough in the filtered signal. Set `feature_value` to the arithmetic mean of these amplitudes.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.183945, + "tolerance": 0.0183945 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_03.json b/physiological_integration/cases/case10/case10_03.json new file mode 100644 index 0000000000000000000000000000000000000000..5a461a52f69e674f65cd1a5575dcc59f55f4a755 --- /dev/null +++ b/physiological_integration/cases/case10/case10_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "From SaO2/SpO2 channel `SaO2`, analyze the half-open interval `[690, 990)` seconds at a sampling rate of `1 Hz`.\n\nUse the raw signal without an additional filter.\nKeep values in [50, 100] and require at least 0.95 valid proportion.\nSet `feature_value` to the arithmetic mean of the retained values.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 95.708424, + "tolerance": 9.5708424 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_04.json b/physiological_integration/cases/case10/case10_04.json new file mode 100644 index 0000000000000000000000000000000000000000..f5f2382131c5db87e101c46226195ba4eb98d68a --- /dev/null +++ b/physiological_integration/cases/case10/case10_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "From SaO2/SpO2 channel `SaO2`, analyze the half-open interval `[900, 1200)` seconds at a sampling rate of `1 Hz`.\n\nUse the raw signal without an additional filter.\nKeep values in [50, 100] and require at least 0.95 valid proportion.\nSet `feature_value` to the minimum of the retained values.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 96.09369, + "tolerance": 9.609369000000001 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_05.json b/physiological_integration/cases/case10/case10_05.json new file mode 100644 index 0000000000000000000000000000000000000000..844f46aa8f82bd83b7d2153676cfbfb1b5368fc4 --- /dev/null +++ b/physiological_integration/cases/case10/case10_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "From SaO2/SpO2 channel `SaO2`, analyze the half-open interval `[1200, 1500)` seconds at a sampling rate of `1 Hz`.\n\nUse the raw signal without an additional filter.\nKeep values in [50, 100] and require at least 0.95 valid proportion.\nSet `feature_value` to the fraction of retained values below 90.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_06.json b/physiological_integration/cases/case10/case10_06.json new file mode 100644 index 0000000000000000000000000000000000000000..60a767f60127f65444fc7d0e823b10be56ade53e --- /dev/null +++ b/physiological_integration/cases/case10/case10_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "From HeartRate channel `H.R.`, analyze the half-open interval `[1500, 1800)` seconds at a sampling rate of `1 Hz`.\n\nUse the raw signal without an additional filter.\nKeep values in [20, 250] and require at least 0.95 valid proportion.\nSet `feature_value` to the arithmetic mean of the retained values.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 74.835101, + "tolerance": 7.4835101 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_07.json b/physiological_integration/cases/case10/case10_07.json new file mode 100644 index 0000000000000000000000000000000000000000..15fca655ddd28f1606824d2940c3695972abfcda --- /dev/null +++ b/physiological_integration/cases/case10/case10_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "From ECG channel `ECG`, analyze the half-open interval `[1800, 1920)` seconds at a sampling rate of `125 Hz`.\n\nApply the declared fourth-order `5–20 Hz` band-pass filter.\nStandardize the filtered signal, take its absolute value, then detect peaks with a minimum distance of 38 samples and a minimum prominence of 1.5.\nFor each pair of consecutive peaks, convert the peak interval to beats per minute as `60 / interval_seconds`. Set `feature_value` to the arithmetic mean of these interval-based heart rates.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 120.691493, + "tolerance": 12.0691493 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_08.json b/physiological_integration/cases/case10/case10_08.json new file mode 100644 index 0000000000000000000000000000000000000000..f18062a74c769bd21eb8d393df4c19daf2304e9c --- /dev/null +++ b/physiological_integration/cases/case10/case10_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "From ECG channel `ECG`, analyze the half-open interval `[2100, 2220)` seconds at a sampling rate of `125 Hz`.\n\nApply the declared fourth-order `5–20 Hz` band-pass filter.\nStandardize the filtered signal, take its absolute value, then detect peaks with a minimum distance of 38 samples and a minimum prominence of 1.5.\nConvert consecutive peak intervals to milliseconds and set `feature_value` to the root mean square of successive differences between those intervals.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 190.559509, + "tolerance": 19.0559509 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_09.json b/physiological_integration/cases/case10/case10_09.json new file mode 100644 index 0000000000000000000000000000000000000000..5832dcab493d196cb370616cf3ceb27574eda55d --- /dev/null +++ b/physiological_integration/cases/case10/case10_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "From EMG channel `EMG`, analyze the half-open interval `[2400, 2460)` seconds at a sampling rate of `125 Hz`.\n\nApply the declared fourth-order `10–45 Hz` band-pass filter.\nSet `feature_value` to the root-mean-square value of the filtered signal.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 6.582936, + "tolerance": 0.6582936 + } + } + ] + } +} diff --git a/physiological_integration/cases/case10/case10_10.json b/physiological_integration/cases/case10/case10_10.json new file mode 100644 index 0000000000000000000000000000000000000000..cef60689979ae29efef1167beffdfee5d349ab06 --- /dev/null +++ b/physiological_integration/cases/case10/case10_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case10_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "From EMG channel `EMG`, analyze the half-open interval `[2700, 2760)` seconds at a sampling rate of `125 Hz`.\n\nApply the declared fourth-order `10–45 Hz` band-pass filter.\nSet `feature_value` to the sum of the absolute differences between successive samples of the filtered signal.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 82118.24828, + "tolerance": 8211.824828 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_01.json b/physiological_integration/cases/case11/case11_01.json new file mode 100644 index 0000000000000000000000000000000000000000..9c83777ce745bfc127003ac99dc856c3dbfe8278 --- /dev/null +++ b/physiological_integration/cases/case11/case11_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[0, 5)` seconds.\n\nFor each sample in `[0, 20)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[0, 20)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.009679826064823573, + "tolerance": 0.0009679826064823573 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_02.json b/physiological_integration/cases/case11/case11_02.json new file mode 100644 index 0000000000000000000000000000000000000000..e02bf43e0b6303266ae2678c885c35fc267fea24 --- /dev/null +++ b/physiological_integration/cases/case11/case11_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[20, 25)` seconds.\n\nFor each sample in `[20, 40)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[20, 40)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.0009058418692967792, + "tolerance": 9.058418692967793e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_03.json b/physiological_integration/cases/case11/case11_03.json new file mode 100644 index 0000000000000000000000000000000000000000..8500c776a8cfe765c69a6ec4f411bb5f2c0cd28d --- /dev/null +++ b/physiological_integration/cases/case11/case11_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[40, 45)` seconds.\n\nFor each sample in `[40, 60)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[40, 60)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.0028910545068996313, + "tolerance": 0.00028910545068996316 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_04.json b/physiological_integration/cases/case11/case11_04.json new file mode 100644 index 0000000000000000000000000000000000000000..3d3cf2960d7cdbfd51a206d59fe428ff57c3dc38 --- /dev/null +++ b/physiological_integration/cases/case11/case11_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[64, 69)` seconds.\n\nFor each sample in `[64, 84)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[64, 84)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.006939466155532795, + "tolerance": 0.0006939466155532795 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_05.json b/physiological_integration/cases/case11/case11_05.json new file mode 100644 index 0000000000000000000000000000000000000000..cfdb2776d3f4e7424fa39cc27d57241c83ad1ce1 --- /dev/null +++ b/physiological_integration/cases/case11/case11_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[80, 85)` seconds.\n\nFor each sample in `[80, 100)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[80, 100)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.0037524504110594452, + "tolerance": 0.00037524504110594457 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_06.json b/physiological_integration/cases/case11/case11_06.json new file mode 100644 index 0000000000000000000000000000000000000000..5120d3e4372008bfa27703406622069e3e5aaa26 --- /dev/null +++ b/physiological_integration/cases/case11/case11_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[100, 105)` seconds.\n\nFor each sample in `[100, 120)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[100, 120)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.007241238834496106, + "tolerance": 0.0007241238834496107 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_07.json b/physiological_integration/cases/case11/case11_07.json new file mode 100644 index 0000000000000000000000000000000000000000..9e11a2e28ba875d336c4036c26b89f96dcd2dd55 --- /dev/null +++ b/physiological_integration/cases/case11/case11_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[120, 125)` seconds.\n\nFor each sample in `[120, 140)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[120, 140)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.0026960320737292012, + "tolerance": 0.00026960320737292015 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_08.json b/physiological_integration/cases/case11/case11_08.json new file mode 100644 index 0000000000000000000000000000000000000000..51cce499bc4cc6bec1746f3b4b0cafab9bf99924 --- /dev/null +++ b/physiological_integration/cases/case11/case11_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[140, 145)` seconds.\n\nFor each sample in `[140, 160)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[140, 160)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.00773230355414958, + "tolerance": 0.0007732303554149581 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_09.json b/physiological_integration/cases/case11/case11_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e6bc419e64f685f8e096d4240f19d0721ac6163a --- /dev/null +++ b/physiological_integration/cases/case11/case11_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[160, 165)` seconds.\n\nFor each sample in `[160, 180)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[160, 180)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.0009014457232930079, + "tolerance": 9.01445723293008e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case11/case11_10.json b/physiological_integration/cases/case11/case11_10.json new file mode 100644 index 0000000000000000000000000000000000000000..315155f1029bfce24dce6812baede4e4acdb4609 --- /dev/null +++ b/physiological_integration/cases/case11/case11_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case11_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf", + "instruction": "Using fNIRS channel `S01D01_760`, calculate the baseline intensity as the median channel value over the half-open interval `[180, 185)` seconds.\n\nFor each sample in `[180, 200)` seconds, calculate optical density as the negative natural logarithm of the sample intensity divided by the baseline intensity. The analysis interval intentionally includes the baseline interval. The channel intensities and baseline median are guaranteed to be finite and positive.\n\nSet `feature_value` to the mean optical density over `[180, 200)` seconds.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.01886077866733824, + "tolerance": 0.001886077866733824 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_01.json b/physiological_integration/cases/case12/case12_01.json new file mode 100644 index 0000000000000000000000000000000000000000..f3104ca3dc5b0ce46ed297798ae96cc9f32d7de2 --- /dev/null +++ b/physiological_integration/cases/case12/case12_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf", + "instruction": "Compute gaze_valid_ratio from [Gaze point X, Gaze point Y] over the half-open interval [0, 8) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the fraction of samples where all gaze channels are finite. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 1.0, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_02.json b/physiological_integration/cases/case12/case12_02.json new file mode 100644 index 0000000000000000000000000000000000000000..8db82e9e35a25ebe118a6af8dbaf4e6b912ed59a --- /dev/null +++ b/physiological_integration/cases/case12/case12_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf", + "instruction": "Compute pupil_left_mean from [Pupil diameter left] over the half-open interval [68, 76) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the mean of the declared pupil channel(s). Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 4.681020572066788, + "tolerance": 0.4681020572066788 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_03.json b/physiological_integration/cases/case12/case12_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c6a52f71b4df42d94d6ce4643f1040b63bda53e4 --- /dev/null +++ b/physiological_integration/cases/case12/case12_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf", + "instruction": "Compute pupil_right_std from [Pupil diameter right] over the half-open interval [40, 48) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the standard deviation of the declared pupil channel(s). Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.3378386534286727, + "tolerance": 0.03378386534286727 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_04.json b/physiological_integration/cases/case12/case12_04.json new file mode 100644 index 0000000000000000000000000000000000000000..5f93f22ccab6e0f80ebc5e315b18f9527d3fe752 --- /dev/null +++ b/physiological_integration/cases/case12/case12_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf", + "instruction": "Compute gaze_dispersion from [Gaze point X, Gaze point Y] over the half-open interval [60, 68) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the sum of the X and Y ranges. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 4163.471519615472, + "tolerance": 416.34715196154724 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_05.json b/physiological_integration/cases/case12/case12_05.json new file mode 100644 index 0000000000000000000000000000000000000000..e6b5eef9c5a8f1692bd93c2b8003053cf732b02a --- /dev/null +++ b/physiological_integration/cases/case12/case12_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf", + "instruction": "Compute gaze_path_length from [Gaze point X, Gaze point Y] over the half-open interval [80, 88) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the sum of Euclidean distances between successive valid gaze points. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 29408.934192309072, + "tolerance": 2940.8934192309075 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_06.json b/physiological_integration/cases/case12/case12_06.json new file mode 100644 index 0000000000000000000000000000000000000000..d956c12cb85c29b313596fd6df9b61cc14be8359 --- /dev/null +++ b/physiological_integration/cases/case12/case12_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf", + "instruction": "Compute gaze_velocity_mean from [Gaze point X, Gaze point Y] over the half-open interval [100, 108) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the mean valid gaze velocity. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 2880.060521746906, + "tolerance": 288.00605217469064 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_07.json b/physiological_integration/cases/case12/case12_07.json new file mode 100644 index 0000000000000000000000000000000000000000..24ff703b9b2babb0ba88627a7ad158b350bc1963 --- /dev/null +++ b/physiological_integration/cases/case12/case12_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf", + "instruction": "Compute binocular_pupil_mean from [Pupil diameter left, Pupil diameter right] over the half-open interval [120, 128) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the mean over valid left and right pupil samples. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 4.665740761697836, + "tolerance": 0.46657407616978364 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_08.json b/physiological_integration/cases/case12/case12_08.json new file mode 100644 index 0000000000000000000000000000000000000000..db735eda2b87b49d33851db070a6c9b023bf09c5 --- /dev/null +++ b/physiological_integration/cases/case12/case12_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf", + "instruction": "Compute pupil_missing_ratio from [Pupil diameter left, Pupil diameter right] over the half-open interval [140, 148) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the fraction of samples missing in the binocular pupil pair. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 0.0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_09.json b/physiological_integration/cases/case12/case12_09.json new file mode 100644 index 0000000000000000000000000000000000000000..11033f80a34308e208a2c41ee43eb42bb4c48989 --- /dev/null +++ b/physiological_integration/cases/case12/case12_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf", + "instruction": "Compute pupil_slope from [Pupil diameter left] over the half-open interval [192, 200) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the linear slope of the declared pupil channel. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": -0.04417999767434465, + "tolerance": 0.004417999767434465 + } + } + ] + } +} diff --git a/physiological_integration/cases/case12/case12_10.json b/physiological_integration/cases/case12/case12_10.json new file mode 100644 index 0000000000000000000000000000000000000000..57699c2af351fc6ee85aa3524a053d202dd7b13a --- /dev/null +++ b/physiological_integration/cases/case12/case12_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case12_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf", + "instruction": "Compute gaze_dispersion from [Gaze point X, Gaze point Y] over the half-open interval [180, 188) seconds. Use the native eye-tracking samples and preserve missing values without imputation. The feature means the sum of the X and Y ranges. Require at least 0.5 valid proportion. Return only feature_value." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 1146.9134789044022, + "tolerance": 114.69134789044023 + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_01.json b/physiological_integration/cases/case13/case13_01.json new file mode 100644 index 0000000000000000000000000000000000000000..813ad330b542c2fa369b8cf003b7c3b81fa8e994 --- /dev/null +++ b/physiological_integration/cases/case13/case13_01.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze EEG channel `Fp1` and GSR channel `GSR1` over [0, 300).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `scl_mean` from the GSR signal as mean skin-conductance level.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_02.json b/physiological_integration/cases/case13/case13_02.json new file mode 100644 index 0000000000000000000000000000000000000000..8fb8aa77b2753911ca87d35006208a795d2b42a8 --- /dev/null +++ b/physiological_integration/cases/case13/case13_02.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze EEG channel `Fz` and GSR channel `GSR1` over [300, 600).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `phasic_auc` from the GSR signal as positive phasic skin-conductance area.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_03.json b/physiological_integration/cases/case13/case13_03.json new file mode 100644 index 0000000000000000000000000000000000000000..12ecd22e64aa38e6e9df3c16ff37b9067240861f --- /dev/null +++ b/physiological_integration/cases/case13/case13_03.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze EEG channel `Cz` and Respiration channel `Resp` over [600, 900).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `theta_alpha_ratio` from the EEG signal as theta-to-alpha power ratio using [4, 8) and [8, 13) Hz. Compute `rate_mean` from the Respiration signal as mean respiratory rate estimated from detected breathing cycles.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_04.json b/physiological_integration/cases/case13/case13_04.json new file mode 100644 index 0000000000000000000000000000000000000000..9d0194d44293517e16cdfbf7806d2ed4582d2df3 --- /dev/null +++ b/physiological_integration/cases/case13/case13_04.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze EEG channel `Pz` and PPG channel `Plet` over [1050, 1350).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `relative_alpha_power` from the EEG signal as alpha power in [8, 13) Hz divided by total power in [0.5, 45) Hz. Compute `rate_mean` from the PPG signal as mean pulse rate estimated from detected pulse cycles.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_05.json b/physiological_integration/cases/case13/case13_05.json new file mode 100644 index 0000000000000000000000000000000000000000..084575a99cb01db432a6dcf063a9ffa8950f2639 --- /dev/null +++ b/physiological_integration/cases/case13/case13_05.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze EEG channel `O1` and Temperature channel `Temp` over [1200, 1500).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `beta_alpha_ratio` from the EEG signal as beta-to-alpha power ratio using [13, 30) and [8, 13) Hz. Compute `mean` from the Temperature signal as mean temperature.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_06.json b/physiological_integration/cases/case13/case13_06.json new file mode 100644 index 0000000000000000000000000000000000000000..aa766ce2a27defa5cb2661f0de3a0ad377dd3455 --- /dev/null +++ b/physiological_integration/cases/case13/case13_06.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze EEG channel `F3` and PPG channel `Plet` over [1500, 1800).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `amplitude_mean` from the PPG signal as mean peak-to-preceding-trough pulse amplitude.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_07.json b/physiological_integration/cases/case13/case13_07.json new file mode 100644 index 0000000000000000000000000000000000000000..8075b97a2528e5331bd4b31b8ba44affc81aeec5 --- /dev/null +++ b/physiological_integration/cases/case13/case13_07.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze EEG channel `FT7` and EOG channel `eog_h` over [24, 104).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `standard_deviation` from the EOG signal as standard deviation of the EOG signal.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_08.json b/physiological_integration/cases/case13/case13_08.json new file mode 100644 index 0000000000000000000000000000000000000000..b004ef266d4242a54a5d72585c9788f665d9ad60 --- /dev/null +++ b/physiological_integration/cases/case13/case13_08.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze EEG channel `OZ` and EOG channel `eog_v` over [80, 160).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `root_mean_square` from the EOG signal as root-mean-square EOG amplitude.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_09.json b/physiological_integration/cases/case13/case13_09.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec345270bb208426d2af7fbba040f4260ac1763 --- /dev/null +++ b/physiological_integration/cases/case13/case13_09.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze EEG channel `POZ` and EOG channel `eog_h` over [160, 240).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_alpha_ratio` from the EEG signal as theta-to-alpha power ratio using [4, 8) and [8, 13) Hz. Compute `blink_amplitude_mean` from the EOG signal as mean absolute amplitude at detected blinks.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_10.json b/physiological_integration/cases/case13/case13_10.json new file mode 100644 index 0000000000000000000000000000000000000000..1a0da4396eb1af701df0877aef30be3ff042e966 --- /dev/null +++ b/physiological_integration/cases/case13/case13_10.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze EEG channel `O2` and EOG channel `eog_v` over [240, 320).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `slow_movement_power` from the EOG signal as EOG power in [0.1, 1.0) Hz.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_11.json b/physiological_integration/cases/case13/case13_11.json new file mode 100644 index 0000000000000000000000000000000000000000..352b6af79a25ba9d2d72802f22e5b27782d46664 --- /dev/null +++ b/physiological_integration/cases/case13/case13_11.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze EEG channel `CP1` and EOG channel `eog_h` over [320, 400).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `relative_alpha_power` from the EEG signal as alpha power in [8, 13) Hz divided by total power in [0.5, 45) Hz. Compute `line_length` from the EOG signal as sum of absolute successive EOG differences.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_12.json b/physiological_integration/cases/case13/case13_12.json new file mode 100644 index 0000000000000000000000000000000000000000..1282e83c12a85c3998a403f648680c54e6419833 --- /dev/null +++ b/physiological_integration/cases/case13/case13_12.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze EEG channel `PZ` and EOG channel `eog_v` over [400, 480).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `beta_alpha_ratio` from the EEG signal as beta-to-alpha power ratio using [13, 30) and [8, 13) Hz. Compute `blink_rate` from the EOG signal as detected blink count per minute.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_13.json b/physiological_integration/cases/case13/case13_13.json new file mode 100644 index 0000000000000000000000000000000000000000..0704dde7bfe0e7b645b7bd705874254cf2ffcc85 --- /dev/null +++ b/physiological_integration/cases/case13/case13_13.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Analyze EEG channel `EEG(sec)` and Respiration channel `THOR RES` over [0, 600).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `delta_power` from the EEG signal as [0.5, 4) Hz band power. Compute `rate_mean` from the Respiration signal as mean respiratory rate estimated from detected breathing cycles.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_14.json b/physiological_integration/cases/case13/case13_14.json new file mode 100644 index 0000000000000000000000000000000000000000..7bc9e7651540337fb104f029008e0048846c051e --- /dev/null +++ b/physiological_integration/cases/case13/case13_14.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Analyze EEG channel `EEG` and Respiration channel `ABDO RES` over [600, 1200).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `amplitude_mean` from the Respiration signal as mean peak-to-preceding-trough respiratory amplitude.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_15.json b/physiological_integration/cases/case13/case13_15.json new file mode 100644 index 0000000000000000000000000000000000000000..3d9b3fa4db9448c3dfa2e92717338d395d090f90 --- /dev/null +++ b/physiological_integration/cases/case13/case13_15.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Analyze EEG channel `EEG(sec)` and SpO2 channel `SaO2` over [1200, 1800).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `sigma_power` from the EEG signal as [12, 16) Hz band power. Compute `mean` from the SpO2 signal as mean SpO2.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_16.json b/physiological_integration/cases/case13/case13_16.json new file mode 100644 index 0000000000000000000000000000000000000000..0243cc2e827778827cceaa19e23d3e04e5be3c2c --- /dev/null +++ b/physiological_integration/cases/case13/case13_16.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Analyze EEG channel `EEG` and heart rate channel `H.R.` over [1800, 2400).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `delta_power` from the EEG signal as [0.5, 4) Hz band power. Compute `mean` from the heart rate signal as mean heart rate.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_17.json b/physiological_integration/cases/case13/case13_17.json new file mode 100644 index 0000000000000000000000000000000000000000..e77edd2bc101623f43b44eefae640abc1bd90d36 --- /dev/null +++ b/physiological_integration/cases/case13/case13_17.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Analyze EEG channel `EEG(sec)` and ECG channel `ECG` over [2400, 3000).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `rmssd_ms` from the ECG signal as RMSSD of detected RR intervals in milliseconds.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_18.json b/physiological_integration/cases/case13/case13_18.json new file mode 100644 index 0000000000000000000000000000000000000000..f879e7e8013f481aae067dd177bcdf1559bfe075 --- /dev/null +++ b/physiological_integration/cases/case13/case13_18.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Analyze EEG channel `EEG` and EMG channel `EMG` over [3000, 3600).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `gamma_power` from the EEG signal as [30, 45) Hz band power. Compute `root_mean_square` from the EMG signal as root-mean-square EMG amplitude.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_19.json b/physiological_integration/cases/case13/case13_19.json new file mode 100644 index 0000000000000000000000000000000000000000..2c6b1a4064202926f01fed014736fc3ece7f2b3f --- /dev/null +++ b/physiological_integration/cases/case13/case13_19.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `FP1` over [20, 220) and fNIRS channels `S01D01_760`, `S01D01_850` over [16.849, 216.849).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `hbo_mean` from the fNIRS signal as mean HbO concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_20.json b/physiological_integration/cases/case13/case13_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0f93ad710289fde443179166caafa64a45f38ac5 --- /dev/null +++ b/physiological_integration/cases/case13/case13_20.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `O2` over [20, 220) and fNIRS channels `S01D01_760`, `S01D01_850` over [18.105, 218.105).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `hbr_mean` from the fNIRS signal as mean HbR concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_21.json b/physiological_integration/cases/case13/case13_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ad48f52dd5e0e9de157ee4e05df0284c3c3b705e --- /dev/null +++ b/physiological_integration/cases/case13/case13_21.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `Cz` over [40, 240) and fNIRS channels `S01D01_760`, `S01D01_850` over [45.168, 245.168).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `hbo_slope` from the fNIRS signal as HbO concentration-change slope against time.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_22.json b/physiological_integration/cases/case13/case13_22.json new file mode 100644 index 0000000000000000000000000000000000000000..f99c765ddb650c494330ddfb5e0d487900db2e30 --- /dev/null +++ b/physiological_integration/cases/case13/case13_22.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `FP1` over [60, 260) and fNIRS channels `S01D01_760`, `S01D01_850` over [62.749, 262.749).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `relative_alpha_power` from the EEG signal as alpha power in [8, 13) Hz divided by total power in [0.5, 45) Hz. Compute `hbr_slope` from the fNIRS signal as HbR concentration-change slope against time.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_23.json b/physiological_integration/cases/case13/case13_23.json new file mode 100644 index 0000000000000000000000000000000000000000..672fc300d03724d63c92d1cc14824140891318a3 --- /dev/null +++ b/physiological_integration/cases/case13/case13_23.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `O2` over [100, 300) and fNIRS channels `S01D01_760`, `S01D01_850` over [82.078, 282.078).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `theta_alpha_ratio` from the EEG signal as theta-to-alpha power ratio using [4, 8) and [8, 13) Hz. Compute `hbo_auc` from the fNIRS signal as positive-part area of HbO concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_24.json b/physiological_integration/cases/case13/case13_24.json new file mode 100644 index 0000000000000000000000000000000000000000..5e4f260c305e20c0d9376915cd4fecb502ddfb04 --- /dev/null +++ b/physiological_integration/cases/case13/case13_24.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `Cz` over [120, 320) and fNIRS channels `S01D01_760`, `S01D01_850` over [116.849, 316.849).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `beta_alpha_ratio` from the EEG signal as beta-to-alpha power ratio using [13, 30) and [8, 13) Hz. Compute `hbr_auc` from the fNIRS signal as positive-part area of HbR concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_25.json b/physiological_integration/cases/case13/case13_25.json new file mode 100644 index 0000000000000000000000000000000000000000..7fe7eec1b1d803fa0e9cc78b349b4f87749e4d92 --- /dev/null +++ b/physiological_integration/cases/case13/case13_25.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FP1` and eye tracking channels `Gaze point X`, `Gaze point Y` over [16, 96).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `gaze_dispersion` from the eye tracking signal as horizontal gaze range plus vertical gaze range from valid samples.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_26.json b/physiological_integration/cases/case13/case13_26.json new file mode 100644 index 0000000000000000000000000000000000000000..c4e2802d8734d548bb872200ca5a84e8904aec3c --- /dev/null +++ b/physiological_integration/cases/case13/case13_26.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FP2` and eye tracking channel `Pupil diameter left` over [80, 160).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `pupil_left_mean` from the eye tracking signal as mean valid positive left-pupil diameter.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "negative" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_27.json b/physiological_integration/cases/case13/case13_27.json new file mode 100644 index 0000000000000000000000000000000000000000..d28bdcc3142a4a2f64500da6c411dfbc2f0163a9 --- /dev/null +++ b/physiological_integration/cases/case13/case13_27.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FZ` and eye tracking channel `Pupil diameter right` over [360, 440).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `pupil_right_std` from the eye tracking signal as standard deviation of valid positive right-pupil diameter.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_28.json b/physiological_integration/cases/case13/case13_28.json new file mode 100644 index 0000000000000000000000000000000000000000..3ac675c2b039eb64c1bb4284f0c89243e19cb505 --- /dev/null +++ b/physiological_integration/cases/case13/case13_28.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `CZ` and eye tracking channels `Gaze point X`, `Gaze point Y` over [240, 320).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_alpha_ratio` from the EEG signal as theta-to-alpha power ratio using [4, 8) and [8, 13) Hz. Compute `gaze_path_length` from the eye tracking signal as total Euclidean path length across consecutive valid gaze samples.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_29.json b/physiological_integration/cases/case13/case13_29.json new file mode 100644 index 0000000000000000000000000000000000000000..23f0c311a8ca080926ab3d3bb0c2dc1d570fab60 --- /dev/null +++ b/physiological_integration/cases/case13/case13_29.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FP1` and eye tracking channels `Pupil diameter left`, `Pupil diameter right` over [320, 400).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `relative_alpha_power` from the EEG signal as alpha power in [8, 13) Hz divided by total power in [0.5, 45) Hz. Compute `binocular_pupil_mean` from the eye tracking signal as mean bilateral pupil diameter over samples valid in both eyes.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_30.json b/physiological_integration/cases/case13/case13_30.json new file mode 100644 index 0000000000000000000000000000000000000000..f84cbbe6f954d56cdb413d8066aa3e53255ac010 --- /dev/null +++ b/physiological_integration/cases/case13/case13_30.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FZ` and eye tracking channel `Pupil diameter left` over [400, 480).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `beta_alpha_ratio` from the EEG signal as beta-to-alpha power ratio using [13, 30) and [8, 13) Hz. Compute `pupil_slope` from the eye tracking signal as left-pupil diameter slope against time.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for multimodal physiological correlation reports.\n\n### TASK\nExtract `correlation_sign`, the reported sign of the correlation between the two\nwindow-level feature sequences.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, numeric correlation values, or extra fields.\n3. The only key must be `correlation_sign`.\n4. The value must be exactly `positive` or `negative`, or null if the sign is\n missing, zero, or unclear.\n\n### OUTPUT TEMPLATE\n{\"correlation_sign\": \"\"}", + "metrics": [ + { + "metric_id": "correlation_sign", + "type": "categorical_check", + "target_key": "correlation_sign", + "weight": 100, + "params": { + "gt_value": "positive" + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_31.json b/physiological_integration/cases/case13/case13_31.json new file mode 100644 index 0000000000000000000000000000000000000000..fc6e1a6b4bb91dfe81b02285e2ab66cfe49cbb56 --- /dev/null +++ b/physiological_integration/cases/case13/case13_31.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze EEG channel `Fp1` and EOG channel `EOG-H` over [0, 300).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `standard_deviation` from the EOG signal as standard deviation of the EOG signal.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `EOG` modality and/or channel(s) `EOG-H` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_32.json b/physiological_integration/cases/case13/case13_32.json new file mode 100644 index 0000000000000000000000000000000000000000..ea6a1ccdd0926ae897c27feaab6eb5dc31450ad1 --- /dev/null +++ b/physiological_integration/cases/case13/case13_32.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze EEG channel `Fz` and fNIRS channels `S03D02_760`, `S03D02_850` over [300, 600).\n\nDivide both streams into aligned, non-overlapping 30-second windows and pair the windows by index.\n\nFor each window, compute `beta_power` from the EEG signal as [13, 30) Hz band power. Compute `hbo_mean` from the fNIRS signal as mean HbO concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `fNIRS` modality and/or channel(s) `S03D02_760, S03D02_850` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_33.json b/physiological_integration/cases/case13/case13_33.json new file mode 100644 index 0000000000000000000000000000000000000000..dec47d1eb8758f1f05fc792b622e7359b6aaa894 --- /dev/null +++ b/physiological_integration/cases/case13/case13_33.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze EEG channel `FT7` and GSR channel `GSR1` over [24, 104).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `scl_mean` from the GSR signal as mean skin-conductance level.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `GSR` modality and/or channel(s) `GSR1` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_34.json b/physiological_integration/cases/case13/case13_34.json new file mode 100644 index 0000000000000000000000000000000000000000..dc4623d4b04e409b4581059ee8b91d9acdaa13b6 --- /dev/null +++ b/physiological_integration/cases/case13/case13_34.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze EEG channel `OZ` and eye tracking channel `PupilLeft` over [80, 160).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `pupil_left_mean` from the eye tracking signal as mean valid positive left-pupil diameter.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `eye tracking` modality and/or channel(s) `PupilLeft` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_35.json b/physiological_integration/cases/case13/case13_35.json new file mode 100644 index 0000000000000000000000000000000000000000..c21ae61c63ed900b63c41f7dd032af05fe09e0c6 --- /dev/null +++ b/physiological_integration/cases/case13/case13_35.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Analyze EEG channel `EEG(sec)` and GSR channel `GSR1` over [0, 600).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `delta_power` from the EEG signal as [0.5, 4) Hz band power. Compute `scl_mean` from the GSR signal as mean skin-conductance level.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `GSR` modality and/or channel(s) `GSR1` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_36.json b/physiological_integration/cases/case13/case13_36.json new file mode 100644 index 0000000000000000000000000000000000000000..34ed12098287e0663b0a8d278bed29199aa83f14 --- /dev/null +++ b/physiological_integration/cases/case13/case13_36.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case13_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Analyze EEG channel `EEG` and eye tracking channel `PupilLeft` over [600, 1200).\n\nDivide both streams into aligned 60-second windows with a 30-second stride, and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `pupil_left_mean` from the eye tracking signal as mean valid positive left-pupil diameter.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `eye tracking` modality and/or channel(s) `PupilLeft` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_37.json b/physiological_integration/cases/case13/case13_37.json new file mode 100644 index 0000000000000000000000000000000000000000..b4cb23ec01b3eabaa36a371aaaf2eea4fac607a5 --- /dev/null +++ b/physiological_integration/cases/case13/case13_37.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `FP1` over [20, 220) and EOG channel `EOG-H` over [16.849, 216.849).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `standard_deviation` from the EOG signal as standard deviation of the EOG signal.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `EOG` modality and/or channel(s) `EOG-H` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_38.json b/physiological_integration/cases/case13/case13_38.json new file mode 100644 index 0000000000000000000000000000000000000000..da1b2f22f506b4ddf85859f59c71d4a6dc4fdc71 --- /dev/null +++ b/physiological_integration/cases/case13/case13_38.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG channel `O2` over [20, 220) and GSR channel `GSR1` over [18.105, 218.105).\n\nDivide both streams into aligned, non-overlapping 20-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `scl_mean` from the GSR signal as mean skin-conductance level.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `GSR` modality and/or channel(s) `GSR1` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_39.json b/physiological_integration/cases/case13/case13_39.json new file mode 100644 index 0000000000000000000000000000000000000000..65fc15103cfb94688afcef22d69c2642b7522920 --- /dev/null +++ b/physiological_integration/cases/case13/case13_39.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FP1` and EOG channel `EOG-H` over [16, 96).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `alpha_power` from the EEG signal as [8, 13) Hz band power. Compute `standard_deviation` from the EOG signal as standard deviation of the EOG signal.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `EOG` modality and/or channel(s) `EOG-H` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case13/case13_40.json b/physiological_integration/cases/case13/case13_40.json new file mode 100644 index 0000000000000000000000000000000000000000..f4c13691b334e0764c8fd0fd6319054a74cb0a83 --- /dev/null +++ b/physiological_integration/cases/case13/case13_40.json @@ -0,0 +1,29 @@ +{ + "meta_info": { + "case_id": "case13_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Analyze EEG channel `FP2` and fNIRS channels `S03D02_760`, `S03D02_850` over [80, 160).\n\nDivide both streams into aligned, non-overlapping 8-second windows and pair the windows by index.\n\nFor each window, compute `theta_power` from the EEG signal as [4, 8) Hz band power. Compute `hbo_mean` from the fNIRS signal as mean HbO concentration change.\n\nKeep windows with finite values for both features. If at least 6 valid pairs remain and both feature sequences have non-zero variance, compute their Pearson correlation. Otherwise, the sign cannot be determined.\n\nReport correlation_sign as `positive` when the correlation is greater than zero and `negative` when it is less than zero. Report null when the correlation cannot be determined or is exactly zero.\n\nReport only correlation_sign." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated from the agent's complete response by a semantic judge.\nReturn an empty JSON object.\n\n### STRICT CONSTRAINTS\n1. Output ONLY a valid empty JSON object.\n2. Do not output Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_modality_channel_handling", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an unavailable modality or channel in the supplied recording.\n\nThe requested `fNIRS` modality and/or channel(s) `S03D02_760, S03D02_850` are not present in the file specified by `data_path`, so the requested window-level feature sequence and `correlation_value` cannot be computed.\n\nA correct response must:\n1. Clearly state that the requested modality or channel is unavailable in the supplied recording.\n2. State or clearly imply that the requested correlation cannot therefore be computed.\n3. Avoid fabricating, estimating, or inferring the missing physiological signal.\n4. Avoid reporting a numerical `correlation_value`, including `0.0`, as though it were a computed result.\n5. Base the conclusion on data availability rather than physiological interpretation, statistical significance, or data quality speculation.\n\nThe exact wording and level of detail may vary. Return {\"status\": true} only when all applicable requirements are satisfied; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key `status`, without explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_01.json b/physiological_integration/cases/case14/case14_01.json new file mode 100644 index 0000000000000000000000000000000000000000..187f5694c0fa1c28f4143e662e6042f6e6bb929b --- /dev/null +++ b/physiological_integration/cases/case14/case14_01.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Using the synchronized event at `233.96875` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `Fp1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [8, 13) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor GSR channel `GSR1`, remove non-finite samples, apply a fourth-order zero-phase Butterworth low-pass filter at 5 Hz, and smooth the filtered signal with a centered 10-second moving average using reflected boundary padding. Compute the mean smoothed GSR value as `scl_mean`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.6328475421208974, + "tolerance": 0.06328475421208975 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -97.17759135443976, + "tolerance": 9.717759135443977 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_02.json b/physiological_integration/cases/case14/case14_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b4e1681cf48b6f1f39b66103a976e37ca53b3184 --- /dev/null +++ b/physiological_integration/cases/case14/case14_02.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Using the synchronized event at `406.316406` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `Fz`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute Hjorth mobility as `sqrt(var(diff(filtered)) / (var(filtered) + 1e-12))`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor GSR channel `GSR1`, remove non-finite samples, apply a fourth-order zero-phase Butterworth low-pass filter at 5 Hz, and smooth the filtered signal with a centered 10-second moving average using reflected boundary padding. Set `phasic = filtered − tonic` and compute its positive-part trapezoidal area as `phasic_auc`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.06018796827332516, + "tolerance": 0.006018796827332517 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -6.227674993282109, + "tolerance": 0.622767499328211 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_03.json b/physiological_integration/cases/case14/case14_03.json new file mode 100644 index 0000000000000000000000000000000000000000..4a6380902da8317d6cf94b5b00eda0d1f078c5cf --- /dev/null +++ b/physiological_integration/cases/case14/case14_03.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Using the synchronized event at `527.511719` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `Cz`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute theta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [4, 8) Hz and [8, 13) Hz with the trapezoidal rule, then set `theta_alpha_ratio = theta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor Respiration channel `Resp`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.05 to 1 Hz. Robust-standardize the filtered signal using median and MAD with scale `1.4826 × MAD`, falling back to standard deviation if necessary. Detect peaks (and troughs for amplitude) with minimum peak distance 2 seconds and prominence 0.5. Compute mean `60 / peak_interval_seconds` as the respiration rate feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.04669452404956287, + "tolerance": 0.004669452404956287 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.6418974821141639, + "tolerance": 0.06418974821141639 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_04.json b/physiological_integration/cases/case14/case14_04.json new file mode 100644 index 0000000000000000000000000000000000000000..1440902c4d1674906cc5fcf8592fb05cb90a00e2 --- /dev/null +++ b/physiological_integration/cases/case14/case14_04.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Using the synchronized event at `725.367188` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `Pz`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` over [8, 13) Hz and `total_power` over [0.5, 45) Hz using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate both bands with the trapezoidal rule, then set `relative_alpha_power = alpha_power / (total_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor PPG channel `Plet`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 8 Hz. Robust-standardize the filtered signal using median and MAD with scale `1.4826 × MAD`, falling back to standard deviation if necessary. Detect peaks (and troughs for amplitude) with minimum peak distance 0.4 seconds and prominence 0.5. Compute mean `60 / peak_interval_seconds` as the PPG rate feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.002452200529618781, + "tolerance": 0.0002452200529618781 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -6.005950519104992, + "tolerance": 0.6005950519104992 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_05.json b/physiological_integration/cases/case14/case14_05.json new file mode 100644 index 0000000000000000000000000000000000000000..f2cb218e420093481a48a7c08ec092ef9713c4d7 --- /dev/null +++ b/physiological_integration/cases/case14/case14_05.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Using the synchronized event at `1134.484375` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `O1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute beta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [13, 30) Hz and [8, 13) Hz with the trapezoidal rule, then set `beta_alpha_ratio = beta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor temperature channel `Temp`, remove non-finite samples, apply a centered 5-second moving average with reflected boundary padding, and Compute the mean as the temperature feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.2726256395895432, + "tolerance": 0.027262563958954322 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.1844180041101815, + "tolerance": 0.01844180041101815 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_06.json b/physiological_integration/cases/case14/case14_06.json new file mode 100644 index 0000000000000000000000000000000000000000..24cd97f4013b17d3606b856366e8a643ad72691d --- /dev/null +++ b/physiological_integration/cases/case14/case14_06.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case14_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Using the synchronized event at `1118.4375` seconds, compare the 30-second pre-event interval with the 30-second post-event interval.\n\nFor EEG channel `F3`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute theta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [4, 8) Hz and [8, 13) Hz with the trapezoidal rule, then set `theta_alpha_ratio = theta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor PPG channel `Plet`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 8 Hz. Robust-standardize the filtered signal using median and MAD with scale `1.4826 × MAD`, falling back to standard deviation if necessary. Detect peaks (and troughs for amplitude) with minimum peak distance 0.4 seconds and prominence 0.5. For each peak, use the nearest preceding trough and average peak-minus-trough amplitude. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.3408395861487309, + "tolerance": 0.03408395861487309 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 1106.010937853187, + "tolerance": 110.60109378531871 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_07.json b/physiological_integration/cases/case14/case14_07.json new file mode 100644 index 0000000000000000000000000000000000000000..aa1a91a0e822a8dcbbac13c970b3c9fb0d29103a --- /dev/null +++ b/physiological_integration/cases/case14/case14_07.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "label_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.npy", + "instruction": "Using the synchronized event at `104` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `FT7`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `theta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [4, 8) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_h`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Compute the standard deviation as the EOG feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -8.48877925114738, + "tolerance": 0.848877925114738 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 20.96107203530933, + "tolerance": 2.096107203530933 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_08.json b/physiological_integration/cases/case14/case14_08.json new file mode 100644 index 0000000000000000000000000000000000000000..118a0f75de0bb959738d731edf4ce91480097ff8 --- /dev/null +++ b/physiological_integration/cases/case14/case14_08.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "label_path": "data/multi/SEEDVIG_02_session-20151106_noon.npy", + "instruction": "Using the synchronized event at `2400` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `OZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [8, 13) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_v`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Compute the root-mean-square amplitude as the EOG feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 3.1009719666187294, + "tolerance": 0.31009719666187296 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -10.951873903800134, + "tolerance": 1.0951873903800136 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_09.json b/physiological_integration/cases/case14/case14_09.json new file mode 100644 index 0000000000000000000000000000000000000000..a962c5c476f6035985a7dd5dbbad4b33cbb84f6d --- /dev/null +++ b/physiological_integration/cases/case14/case14_09.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "label_path": "data/multi/SEEDVIG_03_session-20151024_noon.npy", + "instruction": "Using the synchronized event at `416` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `POZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute theta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [4, 8) Hz and [8, 13) Hz with the trapezoidal rule, then set `theta_alpha_ratio = theta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_h`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Robust-standardize with median/MAD (scale 1.4826), detect positive and negative peaks on the signal and its negation with 0.304-second minimum distance and prominence 2.5. Compute total peak count divided by minutes. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.555814737146681, + "tolerance": 0.055581473714668095 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 4.5, + "tolerance": 0.45 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_10.json b/physiological_integration/cases/case14/case14_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5f8af4d7cf9773c997ecb78a2226c3d46e500393 --- /dev/null +++ b/physiological_integration/cases/case14/case14_10.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "label_path": "data/multi/SEEDVIG_04_session-20151105_noon.npy", + "instruction": "Using the synchronized event at `2208` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `O2`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `beta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [13, 30) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_v`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Estimate Welch PSD with at most 256 samples per segment, integrate the half-open [0.1, 1.0) Hz band, and use that value as the slow-movement power. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 4.484056188789745, + "tolerance": 0.44840561887897457 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 2258.14502150956, + "tolerance": 225.81450215095603 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_11.json b/physiological_integration/cases/case14/case14_11.json new file mode 100644 index 0000000000000000000000000000000000000000..07d3a0d39694bac5278739309499a3b8c70e43c7 --- /dev/null +++ b/physiological_integration/cases/case14/case14_11.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "label_path": "data/multi/SEEDVIG_05_session-20141108_noon.npy", + "instruction": "Using the synchronized event at `592` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `CP1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` over [8, 13) Hz and `total_power` over [0.5, 45) Hz using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate both bands with the trapezoidal rule, then set `relative_alpha_power = alpha_power / (total_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_h`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Compute the sum of absolute successive differences as the EOG feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.06780852074781005, + "tolerance": 0.006780852074781005 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -1455.2208306251468, + "tolerance": 145.5220830625147 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_12.json b/physiological_integration/cases/case14/case14_12.json new file mode 100644 index 0000000000000000000000000000000000000000..2fd8dbde44f083f6998105ab8b69b266a67bdfb7 --- /dev/null +++ b/physiological_integration/cases/case14/case14_12.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "label_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.npy", + "instruction": "Using the synchronized event at `424` seconds, compare the 40-second pre-event interval with the 40-second post-event interval.\n\nFor EEG channel `PZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute beta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [13, 30) Hz and [8, 13) Hz with the trapezoidal rule, then set `beta_alpha_ratio = beta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `eog_v`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Robust-standardize with median/MAD (scale 1.4826), detect positive and negative peaks on the signal and its negation with 0.304-second minimum distance and prominence 2.5. Average absolute filtered amplitude at all detected peaks. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.710158720959857, + "tolerance": 0.0710158720959857 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -39.23392922006451, + "tolerance": 3.9233929220064514 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_13.json b/physiological_integration/cases/case14/case14_13.json new file mode 100644 index 0000000000000000000000000000000000000000..86bc0f68e72f31d1d9c52758e1147a20c1aced5d --- /dev/null +++ b/physiological_integration/cases/case14/case14_13.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Using the synchronized event at `3510.9` seconds, compare the 120-second pre-event interval with the 120-second post-event interval.\n\nFor EEG channel `EEG(sec)`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `delta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [0.5, 4) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor Respiration channel `THOR RES`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.05 to 0.7 Hz. Robust-standardize the filtered signal using median and MAD with scale `1.4826 × MAD`, falling back to standard deviation if necessary. Detect peaks (and troughs for amplitude) with minimum peak distance 2 seconds and prominence 0.5. Compute mean `60 / peak_interval_seconds` as the respiration rate feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -309.00100626772746, + "tolerance": 30.900100626772748 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -1.2830818850833445, + "tolerance": 0.12830818850833445 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_14.json b/physiological_integration/cases/case14/case14_14.json new file mode 100644 index 0000000000000000000000000000000000000000..4e2e62f498ad5831630e926d651e9e20e752a18e --- /dev/null +++ b/physiological_integration/cases/case14/case14_14.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Using the synchronized event at `122` seconds, compare the 120-second pre-event interval with the 120-second post-event interval.\n\nFor EEG channel `EEG`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `theta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [4, 8) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor Respiration channel `ABDO RES`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.05 to 0.7 Hz. Robust-standardize the filtered signal using median and MAD with scale `1.4826 × MAD`, falling back to standard deviation if necessary. Detect peaks (and troughs for amplitude) with minimum peak distance 2 seconds and prominence 0.5. For each peak, use the nearest preceding trough and average peak-minus-trough amplitude. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 2.6426102155757043, + "tolerance": 0.2642610215575704 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.03502606566636754, + "tolerance": 0.003502606566636754 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_15.json b/physiological_integration/cases/case14/case14_15.json new file mode 100644 index 0000000000000000000000000000000000000000..ef6381b25beed0961f9e39b04b711ccc9f31216f --- /dev/null +++ b/physiological_integration/cases/case14/case14_15.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Using the EEG event at `7928.5` seconds and the auxiliary event at `7928.5` seconds, compare the 120-second pre-event interval with the 120-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 30 seconds after its event.\n\nFor EEG channel `EEG(sec)`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `sigma_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [12, 16) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor SpO2 channel `SaO2`, keep finite values in [50, 100] only. If the valid ratio is below 0.95, use 0.0; otherwise compute the minimum as the SpO2 feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.33796364353585195, + "tolerance": 0.0337963643535852 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.7812619211108682, + "tolerance": 0.07812619211108683 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_16.json b/physiological_integration/cases/case14/case14_16.json new file mode 100644 index 0000000000000000000000000000000000000000..b11f9ed2a09f186fa085ba97f3d999bde151665c --- /dev/null +++ b/physiological_integration/cases/case14/case14_16.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Using the synchronized event at `2555` seconds, compare the 120-second pre-event interval with the 120-second post-event interval.\n\nFor EEG channel `EEG`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `beta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [13, 30) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor heart-rate channel `H.R.`, keep finite values in [20, 250] only. If the valid ratio is below 0.95, use 0.0; otherwise compute the mean as the heart-rate feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -18.813959942185168, + "tolerance": 1.8813959942185168 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.1367907733780882, + "tolerance": 0.01367907733780882 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_17.json b/physiological_integration/cases/case14/case14_17.json new file mode 100644 index 0000000000000000000000000000000000000000..c55b11433e85ebf062d7cabea23c526a3c15dffa --- /dev/null +++ b/physiological_integration/cases/case14/case14_17.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Using the synchronized event at `676.3` seconds, compare the 120-second pre-event interval with the 120-second post-event interval.\n\nFor EEG channel `EEG(sec)`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `gamma_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [30, 45) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor ECG channel `ECG`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 5 to 20 Hz. Robust-standardize using median/MAD with standard-deviation fallback, and detect R-like peaks on the absolute z-score with 0.30-second minimum distance and prominence 1.5. Compute mean `60 / RR_interval_seconds`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -1.240068241795114, + "tolerance": 0.12400682417951142 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -10.024146162129526, + "tolerance": 1.0024146162129526 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_18.json b/physiological_integration/cases/case14/case14_18.json new file mode 100644 index 0000000000000000000000000000000000000000..eb1ea69eb4992b0fc77cfce1c34485f9c8ca2819 --- /dev/null +++ b/physiological_integration/cases/case14/case14_18.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case14_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Using the synchronized event at `840` seconds, compare the 120-second pre-event interval with the 120-second post-event interval.\n\nFor EEG channel `EEG`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `delta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [0.5, 4) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EMG channel `EMG`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 10 to 45 Hz. Compute `sqrt(mean(filtered^2))`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 60.94954868222361, + "tolerance": 6.094954868222361 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -2.2677721203603483, + "tolerance": 0.22677721203603485 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_19.json b/physiological_integration/cases/case14/case14_19.json new file mode 100644 index 0000000000000000000000000000000000000000..3ca728f0b2e2f65b6c6c67b851eef5109c692b01 --- /dev/null +++ b/physiological_integration/cases/case14/case14_19.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `115.403` seconds and the auxiliary event at `112.32` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `FP1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `theta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [4, 8) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the mean HbO value. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 11.347700143584689, + "tolerance": 1.134770014358469 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -0.00043705891906556377, + "tolerance": 4.370589190655638e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_20.json b/physiological_integration/cases/case14/case14_20.json new file mode 100644 index 0000000000000000000000000000000000000000..c653d5446fe47272d7197ad4e168425a58c762a6 --- /dev/null +++ b/physiological_integration/cases/case14/case14_20.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `113.026` seconds and the auxiliary event at `111.168` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `O2`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [8, 13) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the mean HbR value. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -43.390835064680324, + "tolerance": 4.339083506468032 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -4.083179701030951e-05, + "tolerance": 4.0831797010309515e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_21.json b/physiological_integration/cases/case14/case14_21.json new file mode 100644 index 0000000000000000000000000000000000000000..6a645117511d6916eb692332d635d07ba18cb8f4 --- /dev/null +++ b/physiological_integration/cases/case14/case14_21.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `110.499` seconds and the auxiliary event at `116.352` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `Cz`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `beta_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [13, 30) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the first-order slope of HbO against seconds. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -10.697518026263108, + "tolerance": 1.0697518026263109 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -6.370960173406577e-05, + "tolerance": 6.370960173406577e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_22.json b/physiological_integration/cases/case14/case14_22.json new file mode 100644 index 0000000000000000000000000000000000000000..bd37a79c17ab597ebdbfad17981e974ed03816fb --- /dev/null +++ b/physiological_integration/cases/case14/case14_22.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `109.709` seconds and the auxiliary event at `113.184` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `FP1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` over [8, 13) Hz and `total_power` over [0.5, 45) Hz using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate both bands with the trapezoidal rule, then set `relative_alpha_power = alpha_power / (total_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the first-order slope of HbR against seconds. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.001328915581376211, + "tolerance": 0.0001328915581376211 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -3.3607911995331226e-06, + "tolerance": 3.360791199533123e-07 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_23.json b/physiological_integration/cases/case14/case14_23.json new file mode 100644 index 0000000000000000000000000000000000000000..09e8a7b5497e86245cc26731bdf13a9ed9983a2a --- /dev/null +++ b/physiological_integration/cases/case14/case14_23.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `129.693` seconds and the auxiliary event at `112.704` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `O2`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute theta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [4, 8) Hz and [8, 13) Hz with the trapezoidal rule, then set `theta_alpha_ratio = theta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the positive-part trapezoidal area of HbO. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.14892765158103283, + "tolerance": 0.014892765158103283 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -0.00038592489418442957, + "tolerance": 3.859248941844296e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_24.json b/physiological_integration/cases/case14/case14_24.json new file mode 100644 index 0000000000000000000000000000000000000000..953ce0482503135b70a0048fa7f059aa52b736ae --- /dev/null +++ b/physiological_integration/cases/case14/case14_24.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case14_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.npy" + ], + "instruction": "Using the EEG event at `184.307` seconds and the auxiliary event at `181.152` seconds, compare the 20-second pre-event interval with the 20-second post-event interval. The EEG post-event interval begins 0 seconds after its event, and the auxiliary post-event interval begins 5 seconds after its event.\n\nFor EEG channel `Cz`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute beta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [13, 30) Hz and [8, 13) Hz with the trapezoidal rule, then set `beta_alpha_ratio = beta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor the fNIRS wavelength pair, use native EDF sampling rates and EDF physical intensity values. Select half-open interval samples by `times = arange(n_samples) / fs`. The selected 760-nm and 850-nm intensities must be finite, positive, non-empty, and non-flatline. Use the first 5 seconds as baseline and compute `OD = -log(I / median(I_baseline))` for each wavelength. Smooth both OD signals with a centered 1-second moving average and reflected padding. Use source-detector distance 3 cm, DPF 6 at both wavelengths, and extinction matrix `[[1.486, 3.843], [2.526, 1.798]]`. Compute `Hb = pinv(A) @ OD`, where `A` is the extinction matrix multiplied by wavelength-specific path lengths; Hb[0] is HbO and Hb[1] is HbR. Use the positive-part trapezoidal area of HbR. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.08446405340234064, + "tolerance": 0.008446405340234065 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.0022241945885829164, + "tolerance": 0.00022241945885829165 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_25.json b/physiological_integration/cases/case14/case14_25.json new file mode 100644 index 0000000000000000000000000000000000000000..4494b1d637afe10e924a367ce5333731522c0af8 --- /dev/null +++ b/physiological_integration/cases/case14/case14_25.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case14_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": [ + "data/multi/SEEDVII_01_session-20221001_1.npy", + "data/multi/SEEDVII_01_session-20221001_1_continuous.npy", + "data/multi/SEEDVII_01_session-20221001_1_events.npy" + ], + "instruction": "Using the synchronized event at `240.064` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `FP1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding. Integrate the [8, 13) Hz band with the trapezoidal rule. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor eye-tracking channels `Gaze point X`, `Gaze point Y`, use the native 250-Hz stream, preserve missing samples, and do not interpolate. A gaze sample is valid only when both Gaze point X and Gaze point Y are finite. If valid ratio is below 0.5, use no value; otherwise compute `(max(valid_x)-min(valid_x)) + (max(valid_y)-min(valid_y))`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 1.0411580860278226e-12, + "tolerance": 1.0411580860278226e-13 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -288.8009944304572, + "tolerance": 28.880099443045722 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_26.json b/physiological_integration/cases/case14/case14_26.json new file mode 100644 index 0000000000000000000000000000000000000000..b62a9c693ae10e5f5f14c86841715621212c45a7 --- /dev/null +++ b/physiological_integration/cases/case14/case14_26.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case14_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": [ + "data/multi/SEEDVII_02_session-20221011_1.npy", + "data/multi/SEEDVII_02_session-20221011_1_continuous.npy", + "data/multi/SEEDVII_02_session-20221011_1_events.npy" + ], + "instruction": "Using the synchronized event at `410.563` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `FP2`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute Hjorth mobility for the filtered epoch and for its first difference, then set Hjorth complexity to the mobility ratio. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor eye-tracking channel `Pupil diameter left`, use the native 250-Hz stream, preserve missing samples, and do not interpolate. Keep finite positive Pupil diameter left samples and compute their mean. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -2.2114906933826397, + "tolerance": 0.22114906933826398 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.08900988581492353, + "tolerance": 0.008900988581492353 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_27.json b/physiological_integration/cases/case14/case14_27.json new file mode 100644 index 0000000000000000000000000000000000000000..46dfb4121501bcc128322a1be71e8885faf95975 --- /dev/null +++ b/physiological_integration/cases/case14/case14_27.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case14_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": [ + "data/multi/SEEDVII_03_session-20221009_1.npy", + "data/multi/SEEDVII_03_session-20221009_1_continuous.npy", + "data/multi/SEEDVII_03_session-20221009_1_events.npy" + ], + "instruction": "Using the synchronized event at `658.876` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `FZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute beta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [13, 30) Hz and [8, 13) Hz with the trapezoidal rule, then set `beta_alpha_ratio = beta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor eye-tracking channels `Pupil diameter left`, `Pupil diameter right`, use the native 250-Hz stream, preserve missing samples, and do not interpolate. Keep samples where both pupil diameters are finite and positive, average the two eyes per sample, and then average across samples. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.2904897870540456, + "tolerance": 0.02904897870540456 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -0.08295411470561209, + "tolerance": 0.008295411470561209 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_28.json b/physiological_integration/cases/case14/case14_28.json new file mode 100644 index 0000000000000000000000000000000000000000..25ff7276c6926b3160bd214f200e96489fd7fd53 --- /dev/null +++ b/physiological_integration/cases/case14/case14_28.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case14_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": [ + "data/multi/SEEDVII_04_session-20221030_1.npy", + "data/multi/SEEDVII_04_session-20221030_1_continuous.npy", + "data/multi/SEEDVII_04_session-20221030_1_events.npy" + ], + "instruction": "Using the synchronized event at `1366.67` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `CZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute theta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [4, 8) Hz and [8, 13) Hz with the trapezoidal rule, then set `theta_alpha_ratio = theta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor eye-tracking channel `Pupil diameter left`, use the native 250-Hz stream, preserve missing samples, and do not interpolate. Keep finite positive Pupil diameter left samples and fit a first-order polynomial against seconds; use its slope. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": -0.08101406719142373, + "tolerance": 0.008101406719142374 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.014077046516058343, + "tolerance": 0.0014077046516058345 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_29.json b/physiological_integration/cases/case14/case14_29.json new file mode 100644 index 0000000000000000000000000000000000000000..8d7f706fa9e5f9fa75c8998626a034a5c07df38b --- /dev/null +++ b/physiological_integration/cases/case14/case14_29.json @@ -0,0 +1,42 @@ +{ + "meta_info": { + "case_id": "case14_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "label_path": [ + "data/multi/SEEDVII_05_session-20220929_1.npy", + "data/multi/SEEDVII_05_session-20220929_1_continuous.npy", + "data/multi/SEEDVII_05_session-20220929_1_events.npy" + ], + "instruction": "Using the synchronized event at `1208.14` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `FP1`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute `alpha_power` over [8, 13) Hz and `total_power` over [0.5, 45) Hz using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate both bands with the trapezoidal rule, then set `relative_alpha_power = alpha_power / (total_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor ECG channel `ECG`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 5 to 20 Hz. Robust-standardize using median/MAD with standard-deviation fallback, and detect R-like peaks on the absolute z-score with 0.30-second minimum distance and prominence 1.5. Compute mean `60 / RR_interval_seconds`. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.030424038155379085, + "tolerance": 0.003042403815537909 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": -12.668575083222322, + "tolerance": 1.2668575083222322 + } + } + ] + } +} diff --git a/physiological_integration/cases/case14/case14_30.json b/physiological_integration/cases/case14/case14_30.json new file mode 100644 index 0000000000000000000000000000000000000000..f879fa36bb07b47b58e40f5a00d0d1b6bb00657f --- /dev/null +++ b/physiological_integration/cases/case14/case14_30.json @@ -0,0 +1,42 @@ +{ + "meta_info": { + "case_id": "case14_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "label_path": [ + "data/multi/SEEDVII_01_session-20221001_1.npy", + "data/multi/SEEDVII_01_session-20221001_1_continuous.npy", + "data/multi/SEEDVII_01_session-20221001_1_events.npy" + ], + "instruction": "Using the synchronized event at `1504.241` seconds, compare the 8-second pre-event interval with the 8-second post-event interval.\n\nFor EEG channel `FZ`, remove non-finite samples, subtract the epoch mean, and apply a fourth-order zero-phase Butterworth band-pass filter from 0.5 to 45 Hz. Compute beta and alpha power using a one-sided Welch spectrum with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and no zero-padding; integrate [13, 30) Hz and [8, 13) Hz with the trapezoidal rule, then set `beta_alpha_ratio = beta_power / (alpha_power + 1e-12)`. Report the post-minus-pre difference as `eeg_response_delta`.\n\nFor EOG channel `VEO`, remove non-finite samples and apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz. Compute the sum of absolute successive differences as the EOG feature. Report the post-minus-pre difference as `auxiliary_response_delta`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data.\n\n### TASK\nExtract the two post-minus-pre response deltas from the agent's report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY `eeg_response_delta` and `auxiliary_response_delta`.\n5. Preserve the declared numeric value types.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_response_delta\": , \"auxiliary_response_delta\": }", + "metrics": [ + { + "metric_id": "eeg_response_delta_accuracy", + "type": "numeric_check", + "target_key": "eeg_response_delta", + "weight": 50, + "params": { + "gt_value": 0.16738162366874976, + "tolerance": 0.016738162366874976 + } + }, + { + "metric_id": "auxiliary_response_delta_accuracy", + "type": "numeric_check", + "target_key": "auxiliary_response_delta", + "weight": 50, + "params": { + "gt_value": 0.00011193825232667107, + "tolerance": 1.1193825232667107e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_01.json b/physiological_integration/cases/case15/case15_01.json new file mode 100644 index 0000000000000000000000000000000000000000..12e96ab354da831aee5a90db489d12eee42500c5 --- /dev/null +++ b/physiological_integration/cases/case15/case15_01.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 16th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.12606984090545392, + "tolerance": 0.012606984090545392 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.07157138097812776, + "tolerance": 0.0071571380978127765 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_increase" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_02.json b/physiological_integration/cases/case15/case15_02.json new file mode 100644 index 0000000000000000000000000000000000000000..13f07f8884efc96a8affa76669e9eff125535017 --- /dev/null +++ b/physiological_integration/cases/case15/case15_02.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 12th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.015448289016550681, + "tolerance": 0.0015448289016550681 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.035295511223921494, + "tolerance": 0.0035295511223921495 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_decrease" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_03.json b/physiological_integration/cases/case15/case15_03.json new file mode 100644 index 0000000000000000000000000000000000000000..8c875a4ff0022ab50af426d320752a664016722c --- /dev/null +++ b/physiological_integration/cases/case15/case15_03.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 40th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.19544988831027554, + "tolerance": 0.019544988831027555 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.24080293620639281, + "tolerance": 0.024080293620639282 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_decrease" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_04.json b/physiological_integration/cases/case15/case15_04.json new file mode 100644 index 0000000000000000000000000000000000000000..deb4e1e87b5e0e972f951876dead2ab984b2dffb --- /dev/null +++ b/physiological_integration/cases/case15/case15_04.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 10th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.08946571493135641, + "tolerance": 0.008946571493135642 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.03675753675864968, + "tolerance": 0.003675753675864968 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_05.json b/physiological_integration/cases/case15/case15_05.json new file mode 100644 index 0000000000000000000000000000000000000000..47ace41e65458b26a4ed94b615ee69b7cdd3a85d --- /dev/null +++ b/physiological_integration/cases/case15/case15_05.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 40th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.020602023706957517, + "tolerance": 0.002060202370695752 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.11146361063808996, + "tolerance": 0.011146361063808997 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_06.json b/physiological_integration/cases/case15/case15_06.json new file mode 100644 index 0000000000000000000000000000000000000000..e2b7a0a945836e959d6d26b4d6eb1d8e6ac9cb3c --- /dev/null +++ b/physiological_integration/cases/case15/case15_06.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 21st trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.016125214984282682, + "tolerance": 0.0016125214984282683 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.01360249495263588, + "tolerance": 0.001360249495263588 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_07.json b/physiological_integration/cases/case15/case15_07.json new file mode 100644 index 0000000000000000000000000000000000000000..fdc3fb19cfc9ea2ee2a084fcdf08fe34414d3e58 --- /dev/null +++ b/physiological_integration/cases/case15/case15_07.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 18th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.0031710892177906694, + "tolerance": 0.00031710892177906695 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.08390901583806709, + "tolerance": 0.008390901583806709 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "no_clear_change" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_08.json b/physiological_integration/cases/case15/case15_08.json new file mode 100644 index 0000000000000000000000000000000000000000..11ab9faf2116207a429b822385181c5eed129dec --- /dev/null +++ b/physiological_integration/cases/case15/case15_08.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 1st trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.06358792388352967, + "tolerance": 0.006358792388352967 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.11223684291649505, + "tolerance": 0.011223684291649506 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_09.json b/physiological_integration/cases/case15/case15_09.json new file mode 100644 index 0000000000000000000000000000000000000000..71d3c28a27864c96852dc06531fa4e89b234a42b --- /dev/null +++ b/physiological_integration/cases/case15/case15_09.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 9th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.6586144748351902, + "tolerance": 0.06586144748351902 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.05205382510921197, + "tolerance": 0.005205382510921197 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_increase" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_10.json b/physiological_integration/cases/case15/case15_10.json new file mode 100644 index 0000000000000000000000000000000000000000..49af020a6b566c6fb930e760fb90c40187a4c436 --- /dev/null +++ b/physiological_integration/cases/case15/case15_10.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 17th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.11566710861583529, + "tolerance": 0.01156671086158353 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.004496361649231062, + "tolerance": 0.0004496361649231062 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "no_clear_change" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_11.json b/physiological_integration/cases/case15/case15_11.json new file mode 100644 index 0000000000000000000000000000000000000000..d00cdb6c4dd05341b4b78c1abd31e69d7efccb1c --- /dev/null +++ b/physiological_integration/cases/case15/case15_11.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 23rd trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.2800348416167587, + "tolerance": 0.02800348416167587 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.1694257584686909, + "tolerance": 0.01694257584686909 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_increase" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_12.json b/physiological_integration/cases/case15/case15_12.json new file mode 100644 index 0000000000000000000000000000000000000000..8317ab7c1088db3bcca80933e05bfc14912577a4 --- /dev/null +++ b/physiological_integration/cases/case15/case15_12.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 31st trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.5311696791274045, + "tolerance": 0.053116967912740454 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.08466577697337385, + "tolerance": 0.008466577697337385 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_13.json b/physiological_integration/cases/case15/case15_13.json new file mode 100644 index 0000000000000000000000000000000000000000..ca895b35bee4e8f2824bd6d22c1010429dfd71b5 --- /dev/null +++ b/physiological_integration/cases/case15/case15_13.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 1st trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.43817037422049243, + "tolerance": 0.043817037422049246 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.02944864604828723, + "tolerance": 0.0029448646048287233 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_14.json b/physiological_integration/cases/case15/case15_14.json new file mode 100644 index 0000000000000000000000000000000000000000..9580c324f31a9ed3bb77821d72c5643e08d0249c --- /dev/null +++ b/physiological_integration/cases/case15/case15_14.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 35th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.13962729315513953, + "tolerance": 0.013962729315513954 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.17349562818763786, + "tolerance": 0.01734956281876379 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_decrease" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_15.json b/physiological_integration/cases/case15/case15_15.json new file mode 100644 index 0000000000000000000000000000000000000000..40502f94713848984fc54bd83ee9940796520b81 --- /dev/null +++ b/physiological_integration/cases/case15/case15_15.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 2nd trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.09707647968631417, + "tolerance": 0.009707647968631418 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.04922651759162733, + "tolerance": 0.004922651759162733 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_16.json b/physiological_integration/cases/case15/case15_16.json new file mode 100644 index 0000000000000000000000000000000000000000..bbf7c3c34c23bd30e968d5e8666097731ab4e388 --- /dev/null +++ b/physiological_integration/cases/case15/case15_16.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 5th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.07936442602457301, + "tolerance": 0.007936442602457301 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.041893473544945437, + "tolerance": 0.0041893473544945435 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_17.json b/physiological_integration/cases/case15/case15_17.json new file mode 100644 index 0000000000000000000000000000000000000000..3d6c2911ac5fcc18caca589cdf6d39232b77219c --- /dev/null +++ b/physiological_integration/cases/case15/case15_17.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 7th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.419934356937343, + "tolerance": 0.0419934356937343 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.29709801555699045, + "tolerance": 0.029709801555699047 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_increase" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_18.json b/physiological_integration/cases/case15/case15_18.json new file mode 100644 index 0000000000000000000000000000000000000000..81ad233cf43bc890d6cda69b1069c28ac9ff15dd --- /dev/null +++ b/physiological_integration/cases/case15/case15_18.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 37th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.4115263873752754, + "tolerance": 0.041152638737527544 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": -0.17333866405331833, + "tolerance": 0.017333866405331834 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_decrease" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_19.json b/physiological_integration/cases/case15/case15_19.json new file mode 100644 index 0000000000000000000000000000000000000000..94a6ae8d2cf92bbad0ebe86a0861df1756fe1076 --- /dev/null +++ b/physiological_integration/cases/case15/case15_19.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 40th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": 0.06658382128812972, + "tolerance": 0.006658382128812972 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.12739400544581514, + "tolerance": 0.012739400544581515 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "both_increase" + } + } + ] + } +} diff --git a/physiological_integration/cases/case15/case15_20.json b/physiological_integration/cases/case15/case15_20.json new file mode 100644 index 0000000000000000000000000000000000000000..973d20a8afa1add15fe5a149cc3ff8f7e2b08466 --- /dev/null +++ b/physiological_integration/cases/case15/case15_20.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case15_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Use the recording's Status marker channel and locate entries with value 4; these are the declared trial-start markers. Counting from the first marker, select the 10th trial-start event. Compare the preceding 5-second baseline with the following 60-second stimulus using [5 seconds before the marker, 60 seconds after it). For EEG channels [Fp1, Fp2, AF3, AF4, F3, F4], calculate mean(log right-alpha power − log left-alpha power) over pairs Fp1–Fp2; AF3–AF4; F3–F4 in 8–14 Hz, then subtract baseline from stimulus. For Temp, calculate the stimulus mean minus the preceding baseline mean. Use neutral thresholds of 0.01 and 0.01; map the two signs to both_increase, both_decrease, opposite_direction, or no_clear_change. Return only eeg_asymmetry_delta, temperature_delta, and response_pattern." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_asymmetry_delta: a finite numeric value or null.\n- temperature_delta: a finite numeric value or null.\n- response_pattern: exactly one of `both_increase`, `both_decrease`, `opposite_direction`, `no_clear_change`, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve numeric values as finite JSON numbers.\n6. `response_pattern` must be exactly one of `both_increase`, `both_decrease`, `opposite_direction`, or `no_clear_change`.\n7. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_asymmetry_delta\": , \"temperature_delta\": , \"response_pattern\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_asymmetry_delta", + "type": "numeric_check", + "target_key": "eeg_asymmetry_delta", + "weight": 40, + "params": { + "gt_value": -0.10338957301450692, + "tolerance": 0.010338957301450694 + } + }, + { + "metric_id": "temperature_delta", + "type": "numeric_check", + "target_key": "temperature_delta", + "weight": 40, + "params": { + "gt_value": 0.11803533631618635, + "tolerance": 0.011803533631618636 + } + }, + { + "metric_id": "response_pattern", + "type": "categorical_check", + "target_key": "response_pattern", + "weight": 20, + "params": { + "gt_value": "opposite_direction" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_01.json b/physiological_integration/cases/case16/case16_01.json new file mode 100644 index 0000000000000000000000000000000000000000..abcf4b2c4f924689d63b3f0b525ca7ab4c19cff0 --- /dev/null +++ b/physiological_integration/cases/case16/case16_01.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_02.json b/physiological_integration/cases/case16/case16_02.json new file mode 100644 index 0000000000000000000000000000000000000000..5db12d5a25c02051b5659626cc722e280ebb9b5f --- /dev/null +++ b/physiological_integration/cases/case16/case16_02.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_03.json b/physiological_integration/cases/case16/case16_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c0ab36ba1b7f1310d2d5bde775109d32b2919fc6 --- /dev/null +++ b/physiological_integration/cases/case16/case16_03.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_04.json b/physiological_integration/cases/case16/case16_04.json new file mode 100644 index 0000000000000000000000000000000000000000..d656202e7f830aa014825f3986a472f7e1cb003a --- /dev/null +++ b/physiological_integration/cases/case16/case16_04.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_05.json b/physiological_integration/cases/case16/case16_05.json new file mode 100644 index 0000000000000000000000000000000000000000..2dc4bab991478d64a1b6ed44ec6e87076350f7bb --- /dev/null +++ b/physiological_integration/cases/case16/case16_05.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_06.json b/physiological_integration/cases/case16/case16_06.json new file mode 100644 index 0000000000000000000000000000000000000000..0eee8f31897a9d87f53ec41b5446c35dc1239214 --- /dev/null +++ b/physiological_integration/cases/case16/case16_06.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_07.json b/physiological_integration/cases/case16/case16_07.json new file mode 100644 index 0000000000000000000000000000000000000000..f527e47f8abb95a53f7178436e5ee5de1996e670 --- /dev/null +++ b/physiological_integration/cases/case16/case16_07.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_08.json b/physiological_integration/cases/case16/case16_08.json new file mode 100644 index 0000000000000000000000000000000000000000..6515cfce57bb6e41550f4d7b0b92fd79d3796b15 --- /dev/null +++ b/physiological_integration/cases/case16/case16_08.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_09.json b/physiological_integration/cases/case16/case16_09.json new file mode 100644 index 0000000000000000000000000000000000000000..4ecd9d37b1c37cf21572689901fa99d76559ef3e --- /dev/null +++ b/physiological_integration/cases/case16/case16_09.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_10.json b/physiological_integration/cases/case16/case16_10.json new file mode 100644 index 0000000000000000000000000000000000000000..3fd6597cd887f18c23dc275b06ea9bc807602e4b --- /dev/null +++ b/physiological_integration/cases/case16/case16_10.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 75.0, + "tolerance": 7.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "higher_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_11.json b/physiological_integration/cases/case16/case16_11.json new file mode 100644 index 0000000000000000000000000000000000000000..72d6e7d8c893bf811f47397ce11aaf2433a1a4e8 --- /dev/null +++ b/physiological_integration/cases/case16/case16_11.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_12.json b/physiological_integration/cases/case16/case16_12.json new file mode 100644 index 0000000000000000000000000000000000000000..e6835dc294d404c85e6c30785e5901ea660b447e --- /dev/null +++ b/physiological_integration/cases/case16/case16_12.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_13.json b/physiological_integration/cases/case16/case16_13.json new file mode 100644 index 0000000000000000000000000000000000000000..7d1a8ffee0dfad8cfb924b3766cbe8955b20cc71 --- /dev/null +++ b/physiological_integration/cases/case16/case16_13.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_14.json b/physiological_integration/cases/case16/case16_14.json new file mode 100644 index 0000000000000000000000000000000000000000..77c5645d90ab316565adb005a10f54eb32fedd90 --- /dev/null +++ b/physiological_integration/cases/case16/case16_14.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_15.json b/physiological_integration/cases/case16/case16_15.json new file mode 100644 index 0000000000000000000000000000000000000000..b521cc0abfff3020e1ce910c06885c7faa64b24e --- /dev/null +++ b/physiological_integration/cases/case16/case16_15.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_16.json b/physiological_integration/cases/case16/case16_16.json new file mode 100644 index 0000000000000000000000000000000000000000..b0a1a8429282858709d00d8b1028d7043067012e --- /dev/null +++ b/physiological_integration/cases/case16/case16_16.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_17.json b/physiological_integration/cases/case16/case16_17.json new file mode 100644 index 0000000000000000000000000000000000000000..5f6d805c478fa9e5e676d701794470523a7e0dcf --- /dev/null +++ b/physiological_integration/cases/case16/case16_17.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_18.json b/physiological_integration/cases/case16/case16_18.json new file mode 100644 index 0000000000000000000000000000000000000000..cefdfd78f4a862a5dddcf8d0b9806630014b3e1a --- /dev/null +++ b/physiological_integration/cases/case16/case16_18.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_19.json b/physiological_integration/cases/case16/case16_19.json new file mode 100644 index 0000000000000000000000000000000000000000..b218c767c2d8791eb0b1a5a0e65174aaec7a7dfe --- /dev/null +++ b/physiological_integration/cases/case16/case16_19.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_20.json b/physiological_integration/cases/case16/case16_20.json new file mode 100644 index 0000000000000000000000000000000000000000..6386d6f6d3fe9425e346d69a38f4bc65cbcc0a6d --- /dev/null +++ b/physiological_integration/cases/case16/case16_20.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_21.json b/physiological_integration/cases/case16/case16_21.json new file mode 100644 index 0000000000000000000000000000000000000000..f8ed65db4525a8a231e3bf4025f0461c9c9597e3 --- /dev/null +++ b/physiological_integration/cases/case16/case16_21.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 58.333333333333336, + "tolerance": 5.833333333333334 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 41.666666666666664, + "tolerance": 4.166666666666667 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_22.json b/physiological_integration/cases/case16/case16_22.json new file mode 100644 index 0000000000000000000000000000000000000000..bb85c69de89ff7fb973e40e7622086bc6626e565 --- /dev/null +++ b/physiological_integration/cases/case16/case16_22.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 91.66666666666667, + "tolerance": 9.166666666666668 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "mixed_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_23.json b/physiological_integration/cases/case16/case16_23.json new file mode 100644 index 0000000000000000000000000000000000000000..fe040ea074a770a531664e9fe5c27ab0527810e5 --- /dev/null +++ b/physiological_integration/cases/case16/case16_23.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_24.json b/physiological_integration/cases/case16/case16_24.json new file mode 100644 index 0000000000000000000000000000000000000000..4d34768d74f0ec85138e169b379f5be0d33e3e3f --- /dev/null +++ b/physiological_integration/cases/case16/case16_24.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_25.json b/physiological_integration/cases/case16/case16_25.json new file mode 100644 index 0000000000000000000000000000000000000000..b97975da7f0c6eda3420af00028ac3dd6d038522 --- /dev/null +++ b/physiological_integration/cases/case16/case16_25.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_26.json b/physiological_integration/cases/case16/case16_26.json new file mode 100644 index 0000000000000000000000000000000000000000..41be38ed2d05ee055346a333665b4f46486647b0 --- /dev/null +++ b/physiological_integration/cases/case16/case16_26.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_27.json b/physiological_integration/cases/case16/case16_27.json new file mode 100644 index 0000000000000000000000000000000000000000..89954197cf68f477996de07e5593ec6ec35095b2 --- /dev/null +++ b/physiological_integration/cases/case16/case16_27.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_28.json b/physiological_integration/cases/case16/case16_28.json new file mode 100644 index 0000000000000000000000000000000000000000..23d191ca0fd8f8bdd025fdec594e9f0b4f1d9f7c --- /dev/null +++ b/physiological_integration/cases/case16/case16_28.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_29.json b/physiological_integration/cases/case16/case16_29.json new file mode 100644 index 0000000000000000000000000000000000000000..16bc51e0db15b928aa005afae92c1c64267a9089 --- /dev/null +++ b/physiological_integration/cases/case16/case16_29.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case16/case16_30.json b/physiological_integration/cases/case16/case16_30.json new file mode 100644 index 0000000000000000000000000000000000000000..e056273f949f09e27ff111fe83ec7fa2c83bccba --- /dev/null +++ b/physiological_integration/cases/case16/case16_30.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case16_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "For target interval `[3436, 3443)`, compare EEG and EOG drowsiness-related features with reference intervals `[0, 7)`, `[1200, 1207)`, `[2400, 2407)`, `[4800, 4807)`, and `[6000, 6007)`.\n\nFor EEG, average theta, alpha, and beta power across all EEG channels and compute `log(theta / (alpha + beta))`. Use theta `[4, 8)`, alpha `[8, 14)`, and beta `[14, 31)` Hz with 2-second Welch segments.\n\nFor `EOG-H` and `EOG-V`, average slow-eye-movement power in `[0.1, 1.0)` Hz using 4-second Welch segments after `0.1–15` Hz filtering.\n\nFor each modality, report the target's midrank percentile among the target and five reference values. Report `higher_drowsiness_evidence` when both percentiles exceed `50`, `lower_drowsiness_evidence` when both are below `50`, and `mixed_evidence` otherwise.\n\nReport only `eeg_percentile_rank`, `eog_percentile_rank`, and `drowsiness_evidence_state`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `eeg_percentile_rank`: a finite number or null.\n- `eog_percentile_rank`: a finite number or null.\n- `drowsiness_evidence_state`: `higher_drowsiness_evidence`, `lower_drowsiness_evidence`, `mixed_evidence`, or null.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the three required keys.\n3. Do not output Markdown, explanations, units, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"eeg_percentile_rank\": , \"eog_percentile_rank\": , \"drowsiness_evidence_state\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_percentile_rank", + "type": "numeric_check", + "target_key": "eeg_percentile_rank", + "weight": 35, + "params": { + "gt_value": 8.333333333333334, + "tolerance": 0.8333333333333335 + } + }, + { + "metric_id": "eog_percentile_rank", + "type": "numeric_check", + "target_key": "eog_percentile_rank", + "weight": 35, + "params": { + "gt_value": 25.0, + "tolerance": 2.5 + } + }, + { + "metric_id": "drowsiness_evidence_state", + "type": "categorical_check", + "target_key": "drowsiness_evidence_state", + "weight": 30, + "params": { + "gt_value": "lower_drowsiness_evidence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_01.json b/physiological_integration/cases/case17/case17_01.json new file mode 100644 index 0000000000000000000000000000000000000000..51f313fc9ebcfd66df516710d94fbc23d2d9646d --- /dev/null +++ b/physiological_integration/cases/case17/case17_01.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[59.350, 83.095, 120.641, 146.429, 212.103, 279.790, 323.770, 398.157]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.6666666666666666, + "tolerance": 0.06666666666666667 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.07098765432098762, + "tolerance": 0.007098765432098763 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.666666666667 and p=0.070987654321. A correct response must describe a negative association of moderate or moderately strong magnitude, while acknowledging that the evidence is uncertain or does not meet the conventional p<0.05 threshold. It must treat the result as a descriptive cross-event association between early EEG beta responses and delayed HbO responses. It must not claim causality, statistical significance, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_02.json b/physiological_integration/cases/case17/case17_02.json new file mode 100644 index 0000000000000000000000000000000000000000..7ffa907d1159a8856052f883e3ec0e28e9d9b122 --- /dev/null +++ b/physiological_integration/cases/case17/case17_02.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 70.968, 119.896, 148.798, 197.741, 226.548, 268.843, 353.380]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.35714285714285715, + "tolerance": 0.03571428571428572 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.38512064355625614, + "tolerance": 0.03851206435562562 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.357142857143 and p=0.385120643556. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_03.json b/physiological_integration/cases/case17/case17_03.json new file mode 100644 index 0000000000000000000000000000000000000000..f4abf0e15f451fceeb3647e03b0e4c39440af772 --- /dev/null +++ b/physiological_integration/cases/case17/case17_03.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 79.927, 128.813, 153.226, 217.687, 266.642, 335.551, 415.672]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.833333333333 and p=0.0101755401235. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_04.json b/physiological_integration/cases/case17/case17_04.json new file mode 100644 index 0000000000000000000000000000000000000000..2271002577889bcdca1579df642bc6e894c46a9c --- /dev/null +++ b/physiological_integration/cases/case17/case17_04.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 191.076, 284.348, 328.812, 360.105, 395.692, 420.099]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.6904761904761905, + "tolerance": 0.06904761904761905 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.05799031816457284, + "tolerance": 0.005799031816457284 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.690476190476 and p=0.0579903181646. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_05.json b/physiological_integration/cases/case17/case17_05.json new file mode 100644 index 0000000000000000000000000000000000000000..c81cea9efcea78d6c8664c6fbe465d3c5acad14e --- /dev/null +++ b/physiological_integration/cases/case17/case17_05.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 135.426, 208.876, 268.843, 293.227, 342.243, 415.672]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.4523809523809524, + "tolerance": 0.045238095238095244 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.2604047674368853, + "tolerance": 0.02604047674368853 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.452380952381 and p=0.260404767437. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_06.json b/physiological_integration/cases/case17/case17_06.json new file mode 100644 index 0000000000000000000000000000000000000000..12f74eaef59cb255a0fdea87b925948a5a768ebc --- /dev/null +++ b/physiological_integration/cases/case17/case17_06.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 119.896, 153.226, 213.260, 275.492, 344.440, 397.934]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.30952380952380953, + "tolerance": 0.030952380952380953 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.4556448907375822, + "tolerance": 0.04556448907375822 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.309523809524 and p=0.455644890738. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_07.json b/physiological_integration/cases/case17/case17_07.json new file mode 100644 index 0000000000000000000000000000000000000000..6c8605e7f7e04511d9f06f643bea5470cfa948c1 --- /dev/null +++ b/physiological_integration/cases/case17/case17_07.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 79.927, 119.896, 148.798, 191.076, 271.023, 335.551, 360.105]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.833333333333 and p=0.0101755401235. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_08.json b/physiological_integration/cases/case17/case17_08.json new file mode 100644 index 0000000000000000000000000000000000000000..53cff5b87279b7fd2e832e32c74be36dd0cee50d --- /dev/null +++ b/physiological_integration/cases/case17/case17_08.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 142.084, 186.685, 215.404, 257.677, 344.400]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.30952380952380953, + "tolerance": 0.030952380952380953 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.4556448907375822, + "tolerance": 0.04556448907375822 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.309523809524 and p=0.455644890738. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_09.json b/physiological_integration/cases/case17/case17_09.json new file mode 100644 index 0000000000000000000000000000000000000000..77175386dbf9b182f80796a37f311c30a59ad3aa --- /dev/null +++ b/physiological_integration/cases/case17/case17_09.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 119.780, 144.267, 193.312, 275.404, 328.854, 359.923]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.9285714285714286, + "tolerance": 0.09285714285714286 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.0008629681828999815, + "tolerance": 8.629681828999815e-05 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.928571428571 and p=0.0008629681829. A correct response must describe a negative association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_10.json b/physiological_integration/cases/case17/case17_10.json new file mode 100644 index 0000000000000000000000000000000000000000..d3327a2e198162527a876f5404a4cab9d57c5e06 --- /dev/null +++ b/physiological_integration/cases/case17/case17_10.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 277.648, 326.591, 359.923, 393.375, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.833333333333 and p=0.0101755401235. A correct response must describe a negative association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_11.json b/physiological_integration/cases/case17/case17_11.json new file mode 100644 index 0000000000000000000000000000000000000000..ef8af2c5f265127be5a5554a49c9ceac045b58fc --- /dev/null +++ b/physiological_integration/cases/case17/case17_11.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 119.780, 186.685, 224.232, 290.959, 335.511, 404.434]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.7142857142857143, + "tolerance": 0.07142857142857144 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.0465282322841673, + "tolerance": 0.0046528232284167295 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.714285714286 and p=0.0465282322842. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_12.json b/physiological_integration/cases/case17/case17_12.json new file mode 100644 index 0000000000000000000000000000000000000000..748e6f7570ff2f67fd6ca9ef33c232e6a8fda86f --- /dev/null +++ b/physiological_integration/cases/case17/case17_12.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 191.114, 224.232, 286.515, 415.618]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.5952380952380952, + "tolerance": 0.05952380952380953 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.11952980613879269, + "tolerance": 0.011952980613879269 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.595238095238 and p=0.119529806139. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_13.json b/physiological_integration/cases/case17/case17_13.json new file mode 100644 index 0000000000000000000000000000000000000000..73b8e49eacf41692af9f7925627b35bd2fede9d1 --- /dev/null +++ b/physiological_integration/cases/case17/case17_13.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 193.312, 217.608, 282.107, 322.169, 404.434]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.3333333333333333, + "tolerance": 0.03333333333333333 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.41975308641975323, + "tolerance": 0.04197530864197532 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.333333333333 and p=0.41975308642. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_14.json b/physiological_integration/cases/case17/case17_14.json new file mode 100644 index 0000000000000000000000000000000000000000..3d0a118307ca1d894bcd6d09bbe42e950cc0a07b --- /dev/null +++ b/physiological_integration/cases/case17/case17_14.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 110.574, 140.948, 182.722, 206.513, 270.159, 403.430]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.38095238095238093, + "tolerance": 0.0380952380952381 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.35181255311756476, + "tolerance": 0.035181255311756476 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.380952380952 and p=0.351812553118. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_15.json b/physiological_integration/cases/case17/case17_15.json new file mode 100644 index 0000000000000000000000000000000000000000..e8f2f88b9ec962649538c87660571816e125c11d --- /dev/null +++ b/physiological_integration/cases/case17/case17_15.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 119.275, 149.566, 184.872, 276.619, 350.925, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8571428571428571, + "tolerance": 0.08571428571428572 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.006530017254715298, + "tolerance": 0.0006530017254715299 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.857142857143 and p=0.00653001725472. A correct response must describe a negative association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_16.json b/physiological_integration/cases/case17/case17_16.json new file mode 100644 index 0000000000000000000000000000000000000000..ccbdd430516b4fff03e3c62e602715b5c129084f --- /dev/null +++ b/physiological_integration/cases/case17/case17_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 117.103, 195.673, 248.383, 314.149, 337.860, 388.440]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.6428571428571429, + "tolerance": 0.0642857142857143 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.08555889101267325, + "tolerance": 0.008555889101267325 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.642857142857 and p=0.0855588910127. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_17.json b/physiological_integration/cases/case17/case17_17.json new file mode 100644 index 0000000000000000000000000000000000000000..dcaa1d7a089416f93a4f2f19f9c28caf859aaead --- /dev/null +++ b/physiological_integration/cases/case17/case17_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 119.275, 208.648, 252.733, 285.229, 331.436, 392.746]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.7619047619047619, + "tolerance": 0.0761904761904762 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.02800493915307187, + "tolerance": 0.002800493915307187 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.761904761905 and p=0.0280049391531. A correct response must describe a negative association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_18.json b/physiological_integration/cases/case17/case17_18.json new file mode 100644 index 0000000000000000000000000000000000000000..244943c2a8dd9c4d59ff3f318d3d6fa6a9184c64 --- /dev/null +++ b/physiological_integration/cases/case17/case17_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 110.574, 149.566, 208.648, 283.077, 314.149, 401.324]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.5714285714285714, + "tolerance": 0.05714285714285714 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.13895995716070692, + "tolerance": 0.013895995716070692 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.571428571429 and p=0.138959957161. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_19.json b/physiological_integration/cases/case17/case17_19.json new file mode 100644 index 0000000000000000000000000000000000000000..78a9ae1ad6e48c1e125dd53e081f31c7f049501f --- /dev/null +++ b/physiological_integration/cases/case17/case17_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.692, 70.868, 110.574, 184.872, 248.383, 285.229, 353.109, 399.158]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.38095238095238093, + "tolerance": 0.0380952380952381 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.35181255311756476, + "tolerance": 0.035181255311756476 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.380952380952 and p=0.351812553118. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_20.json b/physiological_integration/cases/case17/case17_20.json new file mode 100644 index 0000000000000000000000000000000000000000..3c7d947e0d0ab87796235999e208c811cdc5e35b --- /dev/null +++ b/physiological_integration/cases/case17/case17_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.159, 70.131, 129.269, 183.918, 214.256, 247.426, 330.303, 419.806]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.9047619047619048, + "tolerance": 0.09047619047619049 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.0020082755054294673, + "tolerance": 0.00020082755054294673 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.904761904762 and p=0.00200827550543. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_21.json b/physiological_integration/cases/case17/case17_21.json new file mode 100644 index 0000000000000000000000000000000000000000..12572048968e106618e1b0b424ac8a1b603b11f3 --- /dev/null +++ b/physiological_integration/cases/case17/case17_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.159, 65.842, 142.134, 201.232, 262.514, 323.770, 349.747, 389.512]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.404761904762 and p=0.319888641229. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_22.json b/physiological_integration/cases/case17/case17_22.json new file mode 100644 index 0000000000000000000000000000000000000000..65db269f04f7edf47b1ea4c8627473769e00390a --- /dev/null +++ b/physiological_integration/cases/case17/case17_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.159, 70.131, 122.797, 146.429, 212.103, 281.983, 341.076, 393.816]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.809523809524 and p=0.0149026676862. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_23.json b/physiological_integration/cases/case17/case17_23.json new file mode 100644 index 0000000000000000000000000000000000000000..88ae1491ae110af3ca9f631e53655242b8d7e857 --- /dev/null +++ b/physiological_integration/cases/case17/case17_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.159, 65.842, 146.429, 192.510, 258.237, 332.458, 387.327, 419.806]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.38095238095238093, + "tolerance": 0.0380952380952381 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.35181255311756476, + "tolerance": 0.035181255311756476 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.380952380952 and p=0.351812553118. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_24.json b/physiological_integration/cases/case17/case17_24.json new file mode 100644 index 0000000000000000000000000000000000000000..db0c7af7cc527e25e699d120088fb75ddf7e799a --- /dev/null +++ b/physiological_integration/cases/case17/case17_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[42.159, 70.131, 120.641, 212.103, 279.790, 349.747, 389.512, 419.806]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.833333333333 and p=0.0101755401235. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_25.json b/physiological_integration/cases/case17/case17_25.json new file mode 100644 index 0000000000000000000000000000000000000000..fd6b8284c27c35a272ac07d5a968f4da1fbb01a4 --- /dev/null +++ b/physiological_integration/cases/case17/case17_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 136.180, 166.229, 199.299, 235.946, 286.269, 343.074]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.47619047619047616, + "tolerance": 0.047619047619047616 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.23293553465009814, + "tolerance": 0.023293553465009817 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=-0.47619047619 and p=0.23293553465. A correct response must describe a negative association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_26.json b/physiological_integration/cases/case17/case17_26.json new file mode 100644 index 0000000000000000000000000000000000000000..adeaf54350a19dffa98648b749a5125db5915da3 --- /dev/null +++ b/physiological_integration/cases/case17/case17_26.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 142.621, 168.346, 201.466, 269.111, 292.698, 410.495]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.9761904761904762, + "tolerance": 0.09761904761904762 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 3.314396026200147e-05, + "tolerance": 3.3143960262001473e-06 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.97619047619 and p=3.3143960262e-05. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_27.json b/physiological_integration/cases/case17/case17_27.json new file mode 100644 index 0000000000000000000000000000000000000000..cc24ab1097dddb22af0c083c5dde99660088d982 --- /dev/null +++ b/physiological_integration/cases/case17/case17_27.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 142.621, 205.807, 238.095, 286.269, 353.822, 440.598]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.9047619047619048, + "tolerance": 0.09047619047619049 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.0020082755054294673, + "tolerance": 0.00020082755054294673 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.904761904762 and p=0.00200827550543. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_28.json b/physiological_integration/cases/case17/case17_28.json new file mode 100644 index 0000000000000000000000000000000000000000..b1288fb4f1a53707d0f9980defa2d13cad9516b8 --- /dev/null +++ b/physiological_integration/cases/case17/case17_28.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 136.180, 216.548, 269.111, 303.443, 336.629, 423.360]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.833333333333 and p=0.0101755401235. A correct response must describe a positive association of strong magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value is below the conventional 0.05 threshold, but sampling uncertainty must still be acknowledged. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_29.json b/physiological_integration/cases/case17/case17_29.json new file mode 100644 index 0000000000000000000000000000000000000000..6618c96a1cc01b81e518fc4579b9d738689c80ef --- /dev/null +++ b/physiological_integration/cases/case17/case17_29.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 136.180, 168.346, 205.807, 305.615, 368.811, 438.420]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.38095238095238093, + "tolerance": 0.0380952380952381 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.35181255311756476, + "tolerance": 0.035181255311756476 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.380952380952 and p=0.351812553118. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case17/case17_30.json b/physiological_integration/cases/case17/case17_30.json new file mode 100644 index 0000000000000000000000000000000000000000..af2efc89a794bae26bf3c1f1fed7152cf4e33b8b --- /dev/null +++ b/physiological_integration/cases/case17/case17_30.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case17_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Assess event-level EEG–fNIRS covariation across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 94.528, 142.621, 166.229, 216.548, 297.033, 334.441, 438.420]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, compute the baseline-corrected early beta response in [13, 30) Hz from EEG channels F1, FC1, and Cz, using [-5, 0) seconds as baseline and [0, 5) seconds as the response period.\n\nFrom fNIRS pair S06D06, compute the baseline-corrected delayed HbO response over [4, 18) seconds for the same events.\n\nAssess the association between the eight EEG and HbO responses using Spearman’s rank correlation and its two-sided p-value.\n\nReport coupling_rho and coupling_p_value. Briefly interpret the direction, approximate strength, and uncertainty of the cross-event association without making a causal claim." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the reported cross-event EEG--fNIRS association statistics:\n- coupling_rho\n- coupling_p_value\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"coupling_rho\" and \"coupling_p_value\".\n4. coupling_rho must be a finite number in [-1, 1], or null if missing.\n5. coupling_p_value must be a finite number in [0, 1], or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": }", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.30952380952380953, + "tolerance": 0.030952380952380953 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.4556448907375822, + "tolerance": 0.04556448907375822 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of an event-level EEG--fNIRS association across eight repeated task events. The analysis yields Spearman rho=0.309523809524 and p=0.455644890738. A correct response must describe a positive association of moderate magnitude, acknowledge uncertainty, and state that this is descriptive cross-event evidence between early EEG beta and delayed HbO responses. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge uncertainty. It must not claim causality, a confirmed neural--hemodynamic mechanism, or that any single event establishes coupling. Exact wording is not required. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_01.json b/physiological_integration/cases/case18/case18_01.json new file mode 100644 index 0000000000000000000000000000000000000000..0c85a37b4c5151cb4aa1ca44f7c86e938862f7d9 --- /dev/null +++ b/physiological_integration/cases/case18/case18_01.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[14820, 15420)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 14820.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 52.619992244019734, + "tolerance": 2.630999612200987 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 6.012730058490956, + "tolerance": 0.30063650292454785 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_02.json b/physiological_integration/cases/case18/case18_02.json new file mode 100644 index 0000000000000000000000000000000000000000..588100390179d1d0e851a22a99b76f5eb760f75b --- /dev/null +++ b/physiological_integration/cases/case18/case18_02.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[3600, 4200)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 3600.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 21.290447553753637, + "tolerance": 1.064522377687682 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 3.1548995774593798, + "tolerance": 0.157744978872969 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_03.json b/physiological_integration/cases/case18/case18_03.json new file mode 100644 index 0000000000000000000000000000000000000000..09841c27eaa3ea8e984321adade9f117cedd8633 --- /dev/null +++ b/physiological_integration/cases/case18/case18_03.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[4650, 5250)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 4650.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 91.01708447800964, + "tolerance": 4.550854223900482 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 4.297500547944088, + "tolerance": 0.2148750273972044 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_04.json b/physiological_integration/cases/case18/case18_04.json new file mode 100644 index 0000000000000000000000000000000000000000..489506635cbac4947103a6dcff45bafa1e86ea7d --- /dev/null +++ b/physiological_integration/cases/case18/case18_04.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[2040, 2640)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 2040.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 18.787321103171834, + "tolerance": 0.9393660551585917 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 2.902167188254845, + "tolerance": 0.14510835941274225 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_05.json b/physiological_integration/cases/case18/case18_05.json new file mode 100644 index 0000000000000000000000000000000000000000..e02abed689e3a1ac10b3c7ac1c05bc47d068c52c --- /dev/null +++ b/physiological_integration/cases/case18/case18_05.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[300, 900)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 330.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 172.38797224376378, + "tolerance": 8.61939861218819 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 3.4735204862024207, + "tolerance": 0.17367602431012105 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_06.json b/physiological_integration/cases/case18/case18_06.json new file mode 100644 index 0000000000000000000000000000000000000000..ef25864b4381a748ce36567922a52de39f4b980c --- /dev/null +++ b/physiological_integration/cases/case18/case18_06.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[20160, 20760)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 20730.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 54.94583808076977, + "tolerance": 2.7472919040384887 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 0.6550129063111846, + "tolerance": 0.03275064531555923 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_07.json b/physiological_integration/cases/case18/case18_07.json new file mode 100644 index 0000000000000000000000000000000000000000..76dd687e06be9f4af35f337fd66b664dee4d8711 --- /dev/null +++ b/physiological_integration/cases/case18/case18_07.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[4350, 4950)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 4920.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 30.636468599620496, + "tolerance": 1.531823429981025 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 2.622269132815645, + "tolerance": 0.13111345664078225 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_08.json b/physiological_integration/cases/case18/case18_08.json new file mode 100644 index 0000000000000000000000000000000000000000..e964ee602a63e960f7d24b323d54b3219538b410 --- /dev/null +++ b/physiological_integration/cases/case18/case18_08.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[10560, 11160)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 10560.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 102.54426356283344, + "tolerance": 5.127213178141672 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 5.351548036992832, + "tolerance": 0.2675774018496416 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_09.json b/physiological_integration/cases/case18/case18_09.json new file mode 100644 index 0000000000000000000000000000000000000000..8e4856b74744fa7ac4b30581072e362e78bc7731 --- /dev/null +++ b/physiological_integration/cases/case18/case18_09.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[9510, 10110)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 9540.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 12.741539278170661, + "tolerance": 0.6370769639085331 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 3.8423917185163083, + "tolerance": 0.19211958592581543 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_10.json b/physiological_integration/cases/case18/case18_10.json new file mode 100644 index 0000000000000000000000000000000000000000..6fe3d78b4c36493e598f6164ca2706100e9628bb --- /dev/null +++ b/physiological_integration/cases/case18/case18_10.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[1080, 1680)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 1650.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 107.38704294838269, + "tolerance": 5.369352147419135 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 6.151214497178622, + "tolerance": 0.30756072485893116 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_11.json b/physiological_integration/cases/case18/case18_11.json new file mode 100644 index 0000000000000000000000000000000000000000..cb09d239b5c5780a77fba0a25fd4b35ee5c5dec0 --- /dev/null +++ b/physiological_integration/cases/case18/case18_11.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[21600, 22200)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 21600.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 142.61632107847552, + "tolerance": 7.130816053923777 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 1.32977534324757, + "tolerance": 0.0664887671623785 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_12.json b/physiological_integration/cases/case18/case18_12.json new file mode 100644 index 0000000000000000000000000000000000000000..42f032e8bb3dcc74251bdf424d3d70149fce498a --- /dev/null +++ b/physiological_integration/cases/case18/case18_12.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[4980, 5580)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 5040.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 59.83354137886625, + "tolerance": 2.9916770689433125 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 2.401315618069677, + "tolerance": 0.12006578090348385 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_13.json b/physiological_integration/cases/case18/case18_13.json new file mode 100644 index 0000000000000000000000000000000000000000..70d355984701caae9054701d440677570aed475c --- /dev/null +++ b/physiological_integration/cases/case18/case18_13.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[21390, 21990)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 21960.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 95.63167646361495, + "tolerance": 4.781583823180748 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 14.757183948645116, + "tolerance": 0.7378591974322558 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_14.json b/physiological_integration/cases/case18/case18_14.json new file mode 100644 index 0000000000000000000000000000000000000000..848cf5a435badbe6760528cec5bfbbb994caa8e7 --- /dev/null +++ b/physiological_integration/cases/case18/case18_14.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[12840, 13440)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 13350.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 49.10529860056043, + "tolerance": 2.455264930028022 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 1.5683124779598656, + "tolerance": 0.07841562389799328 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_15.json b/physiological_integration/cases/case18/case18_15.json new file mode 100644 index 0000000000000000000000000000000000000000..cecafb6c8ea02ab117654feb0626cd01d8f63f0c --- /dev/null +++ b/physiological_integration/cases/case18/case18_15.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[4950, 5550)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 4950.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 152.15796836794192, + "tolerance": 7.607898418397096 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 5.565779352562454, + "tolerance": 0.2782889676281227 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_16.json b/physiological_integration/cases/case18/case18_16.json new file mode 100644 index 0000000000000000000000000000000000000000..629cbb5b5f63391d3432b45a245f267c2bd8368a --- /dev/null +++ b/physiological_integration/cases/case18/case18_16.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[23280, 23880)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 23280.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 68.5777246234193, + "tolerance": 3.4288862311709654 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 1.6045769324004335, + "tolerance": 0.08022884662002168 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_17.json b/physiological_integration/cases/case18/case18_17.json new file mode 100644 index 0000000000000000000000000000000000000000..628785077fb5d0585bd4ee3af10a629bf6a53e48 --- /dev/null +++ b/physiological_integration/cases/case18/case18_17.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[9960, 10560)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 10140.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 66.03399339680927, + "tolerance": 3.3016996698404633 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 4.375247420431756, + "tolerance": 0.2187623710215878 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_18.json b/physiological_integration/cases/case18/case18_18.json new file mode 100644 index 0000000000000000000000000000000000000000..7c91001547fc2f56fa6f1c2189ccca38132eed22 --- /dev/null +++ b/physiological_integration/cases/case18/case18_18.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[24990, 25590)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 25560.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 65.94789956611751, + "tolerance": 3.297394978305876 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 12.012665557888193, + "tolerance": 0.6006332778944097 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_19.json b/physiological_integration/cases/case18/case18_19.json new file mode 100644 index 0000000000000000000000000000000000000000..19afe68218653aecd62ed4a65fc9a04bdeb5fec5 --- /dev/null +++ b/physiological_integration/cases/case18/case18_19.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[14190, 14790)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 14190.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 14.81633178629045, + "tolerance": 0.7408165893145225 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 4.19527807603473, + "tolerance": 0.20976390380173648 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_20.json b/physiological_integration/cases/case18/case18_20.json new file mode 100644 index 0000000000000000000000000000000000000000..cb0296e072b6b8a87f74c864380471abfda8fb2a --- /dev/null +++ b/physiological_integration/cases/case18/case18_20.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[9000, 9600)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 9000.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 65.37877415959997, + "tolerance": 3.2689387079799985 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 5.6545630849609365, + "tolerance": 0.2827281542480468 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_21.json b/physiological_integration/cases/case18/case18_21.json new file mode 100644 index 0000000000000000000000000000000000000000..5c366967013ffb724d77b82db9a4cfee015fa371 --- /dev/null +++ b/physiological_integration/cases/case18/case18_21.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[24840, 25440)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 24840.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 75.8162257217812, + "tolerance": 3.7908112860890597 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 2.506689104552355, + "tolerance": 0.12533445522761774 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_22.json b/physiological_integration/cases/case18/case18_22.json new file mode 100644 index 0000000000000000000000000000000000000000..880da51ed34f6f758b355f215f867589f0056273 --- /dev/null +++ b/physiological_integration/cases/case18/case18_22.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[12090, 12690)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 12120.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 30.126431303046555, + "tolerance": 1.5063215651523278 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 1.1053166383513213, + "tolerance": 0.055265831917566066 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_23.json b/physiological_integration/cases/case18/case18_23.json new file mode 100644 index 0000000000000000000000000000000000000000..ec4062a17401380ea8805b6c505c15d277e8818c --- /dev/null +++ b/physiological_integration/cases/case18/case18_23.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[26070, 26670)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 26130.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 175.00607201737157, + "tolerance": 8.750303600868579 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 9.512545415670347, + "tolerance": 0.4756272707835174 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_24.json b/physiological_integration/cases/case18/case18_24.json new file mode 100644 index 0000000000000000000000000000000000000000..6a64911ad7917ff4c09fa117433113cafad6e15b --- /dev/null +++ b/physiological_integration/cases/case18/case18_24.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[18570, 19170)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 18960.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 16.55664136388954, + "tolerance": 0.8278320681944771 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 5.026952458219406, + "tolerance": 0.2513476229109703 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_25.json b/physiological_integration/cases/case18/case18_25.json new file mode 100644 index 0000000000000000000000000000000000000000..baade79056dd5ee361ac1da5ca4a3afcc678fd6e --- /dev/null +++ b/physiological_integration/cases/case18/case18_25.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[22350, 22950)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 22920.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 125.11929911597372, + "tolerance": 6.255964955798686 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 4.606810603527698, + "tolerance": 0.23034053017638492 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_26.json b/physiological_integration/cases/case18/case18_26.json new file mode 100644 index 0000000000000000000000000000000000000000..31bbdbd78fb7f943b496b466b0ce31d2e3d1acfa --- /dev/null +++ b/physiological_integration/cases/case18/case18_26.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within the declared interval `[26760, 27360)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 27300.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 168.50420119589174, + "tolerance": 8.425210059794587 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 1.1332604698246338, + "tolerance": 0.056663023491231694 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_27.json b/physiological_integration/cases/case18/case18_27.json new file mode 100644 index 0000000000000000000000000000000000000000..05f3068528540fc9a8c55c8f29f08835462e1449 --- /dev/null +++ b/physiological_integration/cases/case18/case18_27.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within the declared interval `[13200, 13800)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 13770.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 206.41178221931636, + "tolerance": 10.32058911096582 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 3.23760407085688, + "tolerance": 0.16188020354284402 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_28.json b/physiological_integration/cases/case18/case18_28.json new file mode 100644 index 0000000000000000000000000000000000000000..9704317e2d583e4f4cc4519c06b9a7f3940f3e48 --- /dev/null +++ b/physiological_integration/cases/case18/case18_28.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within the declared interval `[30720, 31320)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 31290.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 161.3384221921868, + "tolerance": 8.066921109609341 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 12.020587903247545, + "tolerance": 0.6010293951623773 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_29.json b/physiological_integration/cases/case18/case18_29.json new file mode 100644 index 0000000000000000000000000000000000000000..413149621d0d5803e62633290c71517975758bf2 --- /dev/null +++ b/physiological_integration/cases/case18/case18_29.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within the declared interval `[22380, 22980)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 22770.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 36.50796377715269, + "tolerance": 1.8253981888576345 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 6.1698861568794, + "tolerance": 0.30849430784397003 + } + } + ] + } +} diff --git a/physiological_integration/cases/case18/case18_30.json b/physiological_integration/cases/case18/case18_30.json new file mode 100644 index 0000000000000000000000000000000000000000..ac64caa26dcee06f2fd4d2965f2b54fe4543c65a --- /dev/null +++ b/physiological_integration/cases/case18/case18_30.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case18_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within the declared interval `[24990, 25590)` seconds, divide the recording into public 30-second sleep epochs.\n\nFor each epoch, compute slow-eye-movement power from the two public SHHS EOG channels `EOG(L)` and `EOG(R)` in the half-open frequency band `[0.1, 1.0)` Hz. For each EOG channel, apply a fourth-order zero-phase Butterworth band-pass filter from 0.1 to 15 Hz, then use a 4-second Hann-window Welch PSD with 50% overlap, constant detrending, density scaling, and trapezoidal integration over `[0.1, 1.0)` Hz. The 4-second recipe provides at least two valid frequency bins. Express each epoch's mean two-channel EOG power relative to the median value across the annotated N2 epochs within this declared interval. Select the epoch with the highest EOG power ratio; ties are resolved by the earliest epoch.\n\nFor the selected epoch, compute the RMS amplitude of finite physical samples from channel `EMG` in microvolts and express it relative to the median RMS across the same annotated N2 reference epochs.\n\nReport only `selected_epoch_start_sec`, `eog_power_ratio`, and `emg_rms_ratio`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the selected epoch and its EOG and EMG ratios:\n1. selected_epoch_start_sec\n2. eog_power_ratio\n3. emg_rms_ratio\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY \"selected_epoch_start_sec\", \"eog_power_ratio\", and \"emg_rms_ratio\".\n5. selected_epoch_start_sec must be a finite JSON number representing seconds, or null if missing.\n6. eog_power_ratio must be a finite non-negative JSON number, or null if missing.\n7. emg_rms_ratio must be a finite non-negative JSON number, or null if missing.\n8. Extract values only for the epoch identified as having the highest EOG power ratio.\n9. Do not extract values from other epochs or intermediate calculations.\n\n### OUTPUT TEMPLATE\n{\"selected_epoch_start_sec\": , \"eog_power_ratio\": , \"emg_rms_ratio\": }", + "metrics": [ + { + "metric_id": "selected_epoch_start_sec", + "type": "numeric_check", + "target_key": "selected_epoch_start_sec", + "weight": 30, + "params": { + "gt_value": 25560.0, + "tolerance": 0.001 + } + }, + { + "metric_id": "eog_power_ratio", + "type": "numeric_check", + "target_key": "eog_power_ratio", + "weight": 35, + "params": { + "gt_value": 95.71968678468582, + "tolerance": 4.785984339234291 + } + }, + { + "metric_id": "emg_rms_ratio", + "type": "numeric_check", + "target_key": "emg_rms_ratio", + "weight": 35, + "params": { + "gt_value": 7.392451180450659, + "tolerance": 0.369622559022533 + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_01.json b/physiological_integration/cases/case19/case19_01.json new file mode 100644 index 0000000000000000000000000000000000000000..dd907f3880c46e251660c94efc1d4861647e3027 --- /dev/null +++ b/physiological_integration/cases/case19/case19_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case19_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[15240, 15540)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated only from the agent's full natural-language response. Return an empty JSON object.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid empty JSON object.\n2. DO NOT include Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_stage_comparison_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an interval that contains no valid REM epochs but may contain valid N2 epochs.\n\nBecause REM epochs are absent, the requested REM-versus-N2 EOG and EMG comparison cannot be performed. A correct response must:\n1. Clearly state that no valid REM epochs are available in the declared interval.\n2. Explain that the REM--N2 comparison, including the EOG and EMG stage relations, cannot be determined from this interval.\n3. Avoid reporting or implying rem_higher, rem_lower, or similar for either modality.\n4. Avoid claiming that the multimodal findings are consistent or inconsistent with the typical REM pattern, because the required stage comparison data are absent.\n5. Avoid fabricating numerical values, stage effects, statistical significance, or a clinical diagnosis.\n\nThe wording does not need to match exactly. Statements such as insufficient data, comparison unavailable, or cannot determine are acceptable when explicitly linked to the absence of the required stage epochs.\n\nReturn {\"status\": true} only when the response satisfies these requirements. Otherwise return {\"status\": false}.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_02.json b/physiological_integration/cases/case19/case19_02.json new file mode 100644 index 0000000000000000000000000000000000000000..6440d5d21ffb9eadb2fdc4c7a1a20976a34fd00f --- /dev/null +++ b/physiological_integration/cases/case19/case19_02.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[14610, 15030)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_03.json b/physiological_integration/cases/case19/case19_03.json new file mode 100644 index 0000000000000000000000000000000000000000..423ad1bc5fd4b0c52320fb2459015c83c482191a --- /dev/null +++ b/physiological_integration/cases/case19/case19_03.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[8160, 8700)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_04.json b/physiological_integration/cases/case19/case19_04.json new file mode 100644 index 0000000000000000000000000000000000000000..3f94cc7d746f8f4fae92441ca7398dc2f0774735 --- /dev/null +++ b/physiological_integration/cases/case19/case19_04.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[8310, 8970)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_05.json b/physiological_integration/cases/case19/case19_05.json new file mode 100644 index 0000000000000000000000000000000000000000..b39302e3066edf6162df9ab58fef3b9e72a71058 --- /dev/null +++ b/physiological_integration/cases/case19/case19_05.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[9180, 9960)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_06.json b/physiological_integration/cases/case19/case19_06.json new file mode 100644 index 0000000000000000000000000000000000000000..4a73723a862078f33cbde84e584a46ea3681d6b7 --- /dev/null +++ b/physiological_integration/cases/case19/case19_06.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[3300, 4200)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_07.json b/physiological_integration/cases/case19/case19_07.json new file mode 100644 index 0000000000000000000000000000000000000000..bd7488b3173077e21ee4f133d9821f584624bc3d --- /dev/null +++ b/physiological_integration/cases/case19/case19_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case19_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[26970, 27270)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated only from the agent's full natural-language response. Return an empty JSON object.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid empty JSON object.\n2. DO NOT include Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_stage_comparison_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an interval that contains no valid N2 epochs but may contain valid REM epochs.\n\nBecause N2 epochs are absent, the requested N2-versus-REM EOG and EMG comparison cannot be performed. A correct response must:\n1. Clearly state that no valid N2 epochs are available in the declared interval.\n2. Explain that the N2--REM comparison, including the EOG and EMG stage relations, cannot be determined from this interval.\n3. Avoid reporting or implying rem_higher, rem_lower, or similar for either modality.\n4. Avoid claiming that the multimodal findings are consistent or inconsistent with the typical REM pattern, because the required stage comparison data are absent.\n5. Avoid fabricating numerical values, stage effects, statistical significance, or a clinical diagnosis.\n\nThe wording does not need to match exactly. Statements such as insufficient data, comparison unavailable, or cannot determine are acceptable when explicitly linked to the absence of the required stage epochs.\n\nReturn {\"status\": true} only when the response satisfies these requirements. Otherwise return {\"status\": false}.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_08.json b/physiological_integration/cases/case19/case19_08.json new file mode 100644 index 0000000000000000000000000000000000000000..7cda231d2228177f8adb22fff7be227eb729a6fa --- /dev/null +++ b/physiological_integration/cases/case19/case19_08.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[27690, 28110)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_09.json b/physiological_integration/cases/case19/case19_09.json new file mode 100644 index 0000000000000000000000000000000000000000..de26a1180fe4eea34346146136a5fc0f88598659 --- /dev/null +++ b/physiological_integration/cases/case19/case19_09.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[20220, 20760)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_10.json b/physiological_integration/cases/case19/case19_10.json new file mode 100644 index 0000000000000000000000000000000000000000..f4e3c179ced8fe67949a037f4df1da508964f328 --- /dev/null +++ b/physiological_integration/cases/case19/case19_10.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[16710, 17370)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_11.json b/physiological_integration/cases/case19/case19_11.json new file mode 100644 index 0000000000000000000000000000000000000000..46a9362f29f0b8653155d39e269c0c833d397e81 --- /dev/null +++ b/physiological_integration/cases/case19/case19_11.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[5310, 6090)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_12.json b/physiological_integration/cases/case19/case19_12.json new file mode 100644 index 0000000000000000000000000000000000000000..3badcf3cbb98c33d04176b6633d94b92e1d80b79 --- /dev/null +++ b/physiological_integration/cases/case19/case19_12.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[13110, 14010)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_13.json b/physiological_integration/cases/case19/case19_13.json new file mode 100644 index 0000000000000000000000000000000000000000..643bddb9708ae550a4fe4c94a50b2b4d8fd650c5 --- /dev/null +++ b/physiological_integration/cases/case19/case19_13.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case19_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[14310, 14610)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated only from the agent's full natural-language response. Return an empty JSON object.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid empty JSON object.\n2. DO NOT include Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_stage_comparison_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an interval that contains no valid REM epochs but may contain valid N2 epochs.\n\nBecause REM epochs are absent, the requested REM-versus-N2 EOG and EMG comparison cannot be performed. A correct response must:\n1. Clearly state that no valid REM epochs are available in the declared interval.\n2. Explain that the REM--N2 comparison, including the EOG and EMG stage relations, cannot be determined from this interval.\n3. Avoid reporting or implying rem_higher, rem_lower, or similar for either modality.\n4. Avoid claiming that the multimodal findings are consistent or inconsistent with the typical REM pattern, because the required stage comparison data are absent.\n5. Avoid fabricating numerical values, stage effects, statistical significance, or a clinical diagnosis.\n\nThe wording does not need to match exactly. Statements such as insufficient data, comparison unavailable, or cannot determine are acceptable when explicitly linked to the absence of the required stage epochs.\n\nReturn {\"status\": true} only when the response satisfies these requirements. Otherwise return {\"status\": false}.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_14.json b/physiological_integration/cases/case19/case19_14.json new file mode 100644 index 0000000000000000000000000000000000000000..893b1c54de7580400873c222f84aab441fba3c75 --- /dev/null +++ b/physiological_integration/cases/case19/case19_14.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[9090, 9510)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_15.json b/physiological_integration/cases/case19/case19_15.json new file mode 100644 index 0000000000000000000000000000000000000000..e52f72371faf0e78e52589bca7c761b54a1739a2 --- /dev/null +++ b/physiological_integration/cases/case19/case19_15.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[19830, 20370)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_16.json b/physiological_integration/cases/case19/case19_16.json new file mode 100644 index 0000000000000000000000000000000000000000..3c913d9c3ccfff5d5f1a117f5ae157b402c3fdcd --- /dev/null +++ b/physiological_integration/cases/case19/case19_16.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[25110, 25770)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_17.json b/physiological_integration/cases/case19/case19_17.json new file mode 100644 index 0000000000000000000000000000000000000000..c6b38a90e4493c039749ad03bb90446fc744d109 --- /dev/null +++ b/physiological_integration/cases/case19/case19_17.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[13470, 14250)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_18.json b/physiological_integration/cases/case19/case19_18.json new file mode 100644 index 0000000000000000000000000000000000000000..4e2e69291dcedd4a2f7728323f89d0dd0225f185 --- /dev/null +++ b/physiological_integration/cases/case19/case19_18.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "label_path": "data/multi/SHHS1_01.npy", + "instruction": "Within `[21090, 21990)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_19.json b/physiological_integration/cases/case19/case19_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ce3a7b7d6e4491e111c2b5b1f1108a5d5c61575e --- /dev/null +++ b/physiological_integration/cases/case19/case19_19.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case19_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[3180, 3480)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated only from the agent's full natural-language response. Return an empty JSON object.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid empty JSON object.\n2. DO NOT include Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_stage_comparison_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an interval that contains no valid N2 epochs but may contain valid REM epochs.\n\nBecause N2 epochs are absent, the requested N2-versus-REM EOG and EMG comparison cannot be performed. A correct response must:\n1. Clearly state that no valid N2 epochs are available in the declared interval.\n2. Explain that the N2--REM comparison, including the EOG and EMG stage relations, cannot be determined from this interval.\n3. Avoid reporting or implying rem_higher, rem_lower, or similar for either modality.\n4. Avoid claiming that the multimodal findings are consistent or inconsistent with the typical REM pattern, because the required stage comparison data are absent.\n5. Avoid fabricating numerical values, stage effects, statistical significance, or a clinical diagnosis.\n\nThe wording does not need to match exactly. Statements such as insufficient data, comparison unavailable, or cannot determine are acceptable when explicitly linked to the absence of the required stage epochs.\n\nReturn {\"status\": true} only when the response satisfies these requirements. Otherwise return {\"status\": false}.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_20.json b/physiological_integration/cases/case19/case19_20.json new file mode 100644 index 0000000000000000000000000000000000000000..eeda3dad0b64fbd7c0ca5ff40f7ad6f77224fcb6 --- /dev/null +++ b/physiological_integration/cases/case19/case19_20.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[8400, 8820)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_21.json b/physiological_integration/cases/case19/case19_21.json new file mode 100644 index 0000000000000000000000000000000000000000..46a2e395056b7e30ae976e62c42ed190c43384dc --- /dev/null +++ b/physiological_integration/cases/case19/case19_21.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "label_path": "data/multi/SHHS1_02.npy", + "instruction": "Within `[8430, 8970)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_22.json b/physiological_integration/cases/case19/case19_22.json new file mode 100644 index 0000000000000000000000000000000000000000..5fbe9b867f5a31c4d6ee5d47cd9cf8e8f1d42101 --- /dev/null +++ b/physiological_integration/cases/case19/case19_22.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[15450, 16110)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_23.json b/physiological_integration/cases/case19/case19_23.json new file mode 100644 index 0000000000000000000000000000000000000000..0edaf9ba7072aa05e6eb4451009ac129db42b95b --- /dev/null +++ b/physiological_integration/cases/case19/case19_23.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[26280, 27060)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_24.json b/physiological_integration/cases/case19/case19_24.json new file mode 100644 index 0000000000000000000000000000000000000000..687a1877c8c357e4db7515c9ab5849041c72ad2f --- /dev/null +++ b/physiological_integration/cases/case19/case19_24.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "label_path": "data/multi/SHHS1_03.npy", + "instruction": "Within `[27390, 28290)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_25.json b/physiological_integration/cases/case19/case19_25.json new file mode 100644 index 0000000000000000000000000000000000000000..ad53121077e70580b12d0ed3b400b377fc5d90c7 --- /dev/null +++ b/physiological_integration/cases/case19/case19_25.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case19_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[14250, 14550)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a strict JSON formatting engine.\n\n### TASK\nThis case is evaluated only from the agent's full natural-language response. Return an empty JSON object.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid empty JSON object.\n2. DO NOT include Markdown, explanations, or additional fields.\n\n### OUTPUT TEMPLATE\n{}", + "metrics": [ + { + "metric_id": "missing_stage_comparison_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "Judge whether the response correctly handles an interval that contains no valid REM epochs but may contain valid N2 epochs.\n\nBecause REM epochs are absent, the requested REM-versus-N2 EOG and EMG comparison cannot be performed. A correct response must:\n1. Clearly state that no valid REM epochs are available in the declared interval.\n2. Explain that the REM--N2 comparison, including the EOG and EMG stage relations, cannot be determined from this interval.\n3. Avoid reporting or implying rem_higher, rem_lower, or similar for either modality.\n4. Avoid claiming that the multimodal findings are consistent or inconsistent with the typical REM pattern, because the required stage comparison data are absent.\n5. Avoid fabricating numerical values, stage effects, statistical significance, or a clinical diagnosis.\n\nThe wording does not need to match exactly. Statements such as insufficient data, comparison unavailable, or cannot determine are acceptable when explicitly linked to the absence of the required stage epochs.\n\nReturn {\"status\": true} only when the response satisfies these requirements. Otherwise return {\"status\": false}.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_26.json b/physiological_integration/cases/case19/case19_26.json new file mode 100644 index 0000000000000000000000000000000000000000..373fb7374a99a95a30f853f58f8c7762814cee3a --- /dev/null +++ b/physiological_integration/cases/case19/case19_26.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[16920, 17340)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_27.json b/physiological_integration/cases/case19/case19_27.json new file mode 100644 index 0000000000000000000000000000000000000000..fabe01f0a6a69902e651c9fb17cfa7d2b37fb0eb --- /dev/null +++ b/physiological_integration/cases/case19/case19_27.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "label_path": "data/multi/SHHS1_04.npy", + "instruction": "Within `[17910, 18450)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_higher" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_28.json b/physiological_integration/cases/case19/case19_28.json new file mode 100644 index 0000000000000000000000000000000000000000..f5f738882431be81326a463ebe6c8cfd264a262e --- /dev/null +++ b/physiological_integration/cases/case19/case19_28.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[9090, 9750)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_29.json b/physiological_integration/cases/case19/case19_29.json new file mode 100644 index 0000000000000000000000000000000000000000..950b622c51f6b3d5b1bc0bdf27c18806b30de5c2 --- /dev/null +++ b/physiological_integration/cases/case19/case19_29.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[19860, 20640)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case19/case19_30.json b/physiological_integration/cases/case19/case19_30.json new file mode 100644 index 0000000000000000000000000000000000000000..84f203da7c68d6ed601d70edb78b89844e055ead --- /dev/null +++ b/physiological_integration/cases/case19/case19_30.json @@ -0,0 +1,45 @@ +{ + "meta_info": { + "case_id": "case19_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "label_path": "data/multi/SHHS1_05.npy", + "instruction": "Within `[9000, 9900)` seconds, compare REM sleep with N2 sleep using the public 30-second sleep-stage labels. Label `4` denotes REM, and label `2` denotes N2. Use only complete epochs with valid `EOG(L)`, `EOG(R)`, and `EMG` signals.\n\nFor each epoch, compute demeaned RMS activity from the two EOG channels and `EMG`, averaging the two EOG values. Compare the median REM and N2 values for each modality.\n\nReport `rem_higher` when the REM median is at least 10% higher than the N2 median, `rem_lower` when it is at least 10% lower, and `similar` otherwise.\n\nReport `eog_stage_relation` and `emg_stage_relation`. Then briefly interpret whether the combined EOG and EMG findings are consistent with the typical REM pattern of increased eye-movement activity and reduced muscle tone. Treat the findings as a descriptive within-record comparison, not as a statistical significance test or clinical diagnosis." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep neurophysiology reports. Your sole task is to extract the two structured stage-relation fields from the agent's response and return one JSON object. Ignore any natural-language multimodal interpretation outside these fields.\n\n### TASK\nExtract the following fields:\n- eog_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n- emg_stage_relation: one of \"rem_higher\", \"rem_lower\", or \"similar\"; use null if missing or unclear.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"eog_stage_relation\" and \"emg_stage_relation\".\n5. Each non-null value must be exactly one of \"rem_higher\", \"rem_lower\", or \"similar\".\n6. Use null when a field is missing, ambiguous, or assigned an unsupported value.\n7. Do not extract the natural-language multimodal interpretation.\n\n### OUTPUT TEMPLATE\n{\"eog_stage_relation\": \"rem_higher\", \"emg_stage_relation\": \"rem_lower\"}", + "metrics": [ + { + "metric_id": "eog_stage_relation", + "type": "categorical_check", + "target_key": "eog_stage_relation", + "weight": 30, + "params": { + "gt_value": "rem_lower" + } + }, + { + "metric_id": "emg_stage_relation", + "type": "categorical_check", + "target_key": "emg_stage_relation", + "weight": 30, + "params": { + "gt_value": "similar" + } + }, + { + "metric_id": "multimodal_rem_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the combined EOG and EMG findings in the response. Do not score the exact correctness of the structured eog_stage_relation and emg_stage_relation fields; those are evaluated separately. Instead, determine whether the interpretation is consistent with the relation values that the response itself reports.\n\nThe typical REM pattern is increased eye-movement activity and reduced muscle tone relative to N2. Therefore:\n- eog_stage_relation=rem_higher and emg_stage_relation=rem_lower should be interpreted as consistent with the typical multimodal REM pattern.\n- If exactly one modality shows its expected REM direction and the other is similar, the interpretation should describe partial support from that modality.\n- If both modalities are similar, the interpretation should state that there is no clear multimodal REM--N2 difference.\n- eog_stage_relation=rem_lower or emg_stage_relation=rem_higher indicates evidence opposite to the typical REM direction. The interpretation should describe the pattern as discordant, atypical, inconsistent, mixed, or contradictory as appropriate.\n- If one modality shows the expected direction and the other shows the opposite direction, the interpretation must describe mixed or contradictory evidence rather than merely partial support.\n\nThe response must contain an actual combined interpretation, not only repeat the two structured values. It must not claim statistical significance because this task is a descriptive comparison. It must not make a clinical diagnosis or infer a sleep disorder. Minor wording differences are acceptable.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_01.json b/physiological_integration/cases/case2/case2_01.json new file mode 100644 index 0000000000000000000000000000000000000000..3abf52f734a49c78a2832bcb1e6933838e1d0b8f --- /dev/null +++ b/physiological_integration/cases/case2/case2_01.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Count analyzable physiological channels for modality EEG and modality GSR/EDA in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 32, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_02.json b/physiological_integration/cases/case2/case2_02.json new file mode 100644 index 0000000000000000000000000000000000000000..3cc0d57b4295253141a6a09b018fb5411d49626d --- /dev/null +++ b/physiological_integration/cases/case2/case2_02.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Count analyzable physiological channels for modality Respiration and modality PPG/BVP in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_03.json b/physiological_integration/cases/case2/case2_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c0efc15c1cb2b363def387111172b57bd2d0086f --- /dev/null +++ b/physiological_integration/cases/case2/case2_03.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Count analyzable physiological channels for modality Temperature and modality ECG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_04.json b/physiological_integration/cases/case2/case2_04.json new file mode 100644 index 0000000000000000000000000000000000000000..338c73956e4329fd71632a58feeee14e20a2ed6d --- /dev/null +++ b/physiological_integration/cases/case2/case2_04.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Count analyzable physiological channels for modality EEG and modality fNIRS in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 32, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_05.json b/physiological_integration/cases/case2/case2_05.json new file mode 100644 index 0000000000000000000000000000000000000000..a873f9c2766af12b72f5e8499f99669b961cc3d6 --- /dev/null +++ b/physiological_integration/cases/case2/case2_05.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Count analyzable physiological channels for modality GSR/EDA and modality SaO2/SpO2 in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_06.json b/physiological_integration/cases/case2/case2_06.json new file mode 100644 index 0000000000000000000000000000000000000000..518ed4517550031184f06c46215b10c2204a2069 --- /dev/null +++ b/physiological_integration/cases/case2/case2_06.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Count analyzable physiological channels for modality EEG and modality EOG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 17, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_07.json b/physiological_integration/cases/case2/case2_07.json new file mode 100644 index 0000000000000000000000000000000000000000..842195d38cd3e3c9e41d594f4ed64d7a9e914cdd --- /dev/null +++ b/physiological_integration/cases/case2/case2_07.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Count analyzable physiological channels for modality EOG and modality GSR/EDA in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_08.json b/physiological_integration/cases/case2/case2_08.json new file mode 100644 index 0000000000000000000000000000000000000000..3b9eb6f4478e1765f92284bb514d5151a4a09148 --- /dev/null +++ b/physiological_integration/cases/case2/case2_08.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Count analyzable physiological channels for modality EEG and modality ECG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 17, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_09.json b/physiological_integration/cases/case2/case2_09.json new file mode 100644 index 0000000000000000000000000000000000000000..bd9d5a8c68ec998cb74f57bfad94cd33473ba59a --- /dev/null +++ b/physiological_integration/cases/case2/case2_09.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Count analyzable physiological channels for modality EOG and modality fNIRS in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_10.json b/physiological_integration/cases/case2/case2_10.json new file mode 100644 index 0000000000000000000000000000000000000000..b7ccbfe06df58c17607e9fdcb7843c59124af473 --- /dev/null +++ b/physiological_integration/cases/case2/case2_10.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Count analyzable physiological channels for modality EEG and modality Respiration in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 17, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_11.json b/physiological_integration/cases/case2/case2_11.json new file mode 100644 index 0000000000000000000000000000000000000000..6a50231ab093d48a3222c03952ad4b1d48c366a0 --- /dev/null +++ b/physiological_integration/cases/case2/case2_11.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Count analyzable physiological channels for modality EEG and modality EOG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_12.json b/physiological_integration/cases/case2/case2_12.json new file mode 100644 index 0000000000000000000000000000000000000000..b0fe6dc432b5d17952b6672515eb43892d7ce480 --- /dev/null +++ b/physiological_integration/cases/case2/case2_12.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Count analyzable physiological channels for modality Respiration and modality SaO2/SpO2 in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 3, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_13.json b/physiological_integration/cases/case2/case2_13.json new file mode 100644 index 0000000000000000000000000000000000000000..22ed97965a181a23380052d4d7725d9f7a075ae8 --- /dev/null +++ b/physiological_integration/cases/case2/case2_13.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Count analyzable physiological channels for modality ECG and modality HeartRate in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_14.json b/physiological_integration/cases/case2/case2_14.json new file mode 100644 index 0000000000000000000000000000000000000000..12f5d9b6ac508d2679ad96daff90a0074ee770fe --- /dev/null +++ b/physiological_integration/cases/case2/case2_14.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Count analyzable physiological channels for modality EMG and modality GSR/EDA in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_15.json b/physiological_integration/cases/case2/case2_15.json new file mode 100644 index 0000000000000000000000000000000000000000..d729f59d1d5cc05f1488909742be881d872ae74d --- /dev/null +++ b/physiological_integration/cases/case2/case2_15.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case02_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Count analyzable physiological channels for modality EEG and modality fNIRS in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_16.json b/physiological_integration/cases/case2/case2_16.json new file mode 100644 index 0000000000000000000000000000000000000000..d9ea6dd97a58f536e4cd05dee4914f75ac9f8522 --- /dev/null +++ b/physiological_integration/cases/case2/case2_16.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Count analyzable physiological channels for modality EEG and modality EOG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 28, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_17.json b/physiological_integration/cases/case2/case2_17.json new file mode 100644 index 0000000000000000000000000000000000000000..c2b30f25ba16b879170c69372a2e92c5037d2380 --- /dev/null +++ b/physiological_integration/cases/case2/case2_17.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Count analyzable physiological channels for modality fNIRS and modality EEG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 72, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 28, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_18.json b/physiological_integration/cases/case2/case2_18.json new file mode 100644 index 0000000000000000000000000000000000000000..b77bc3b0ab88b0cec28b092a05b7c59e600402c5 --- /dev/null +++ b/physiological_integration/cases/case2/case2_18.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Count analyzable physiological channels for modality EOG and modality ECG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_19.json b/physiological_integration/cases/case2/case2_19.json new file mode 100644 index 0000000000000000000000000000000000000000..95b152a03affa5bc32fec0e232f829817402066d --- /dev/null +++ b/physiological_integration/cases/case2/case2_19.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Count analyzable physiological channels for modality fNIRS and modality SaO2/SpO2 in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 72, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_20.json b/physiological_integration/cases/case2/case2_20.json new file mode 100644 index 0000000000000000000000000000000000000000..64c1c24a80b758875f2cfe00980326a4efc95938 --- /dev/null +++ b/physiological_integration/cases/case2/case2_20.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Count analyzable physiological channels for modality EEG and modality Respiration in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 28, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_21.json b/physiological_integration/cases/case2/case2_21.json new file mode 100644 index 0000000000000000000000000000000000000000..8cebd387a77e6e08d484ce2f11916502a369c2f5 --- /dev/null +++ b/physiological_integration/cases/case2/case2_21.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Count analyzable physiological channels for modality EEG and modality EyeTracking in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 62, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 4, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_22.json b/physiological_integration/cases/case2/case2_22.json new file mode 100644 index 0000000000000000000000000000000000000000..e25d2e2c2908d60423a143105417874ae536b092 --- /dev/null +++ b/physiological_integration/cases/case2/case2_22.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Count analyzable physiological channels for modality EOG and modality ECG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_23.json b/physiological_integration/cases/case2/case2_23.json new file mode 100644 index 0000000000000000000000000000000000000000..aaea4a31f5fca24594e13e2feb6186078afd8d62 --- /dev/null +++ b/physiological_integration/cases/case2/case2_23.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Count analyzable physiological channels for modality EyeTracking and modality fNIRS in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 4, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_24.json b/physiological_integration/cases/case2/case2_24.json new file mode 100644 index 0000000000000000000000000000000000000000..a52037b897ba0d8e6863f323992959ea9ec1f86f --- /dev/null +++ b/physiological_integration/cases/case2/case2_24.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "instruction": "Count analyzable physiological channels for modality ECG and modality Respiration in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 1, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 0, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case2/case2_25.json b/physiological_integration/cases/case2/case2_25.json new file mode 100644 index 0000000000000000000000000000000000000000..88943cb9a5f61adce5cae78c3859f87cabd9a82e --- /dev/null +++ b/physiological_integration/cases/case2/case2_25.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case02_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "instruction": "Count analyzable physiological channels for modality EEG and modality EOG in the provided recording. Do not count markers, status, labels, timestamps, metadata, or trial fields. Return only modality_1_channel_count and modality_2_channel_count." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- modality_1_channel_count: Modality 1 channel count.\n- modality_2_channel_count: Modality 2 channel count.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"modality_1_channel_count\": , \"modality_2_channel_count\": }", + "metrics": [ + { + "metric_id": "modality_1_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_1_channel_count", + "weight": 50, + "params": { + "gt_value": 62, + "tolerance": 0.0 + } + }, + { + "metric_id": "modality_2_channel_count_accuracy", + "type": "numeric_check", + "target_key": "modality_2_channel_count", + "weight": 50, + "params": { + "gt_value": 2, + "tolerance": 0.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_01.json b/physiological_integration/cases/case20/case20_01.json new file mode 100644 index 0000000000000000000000000000000000000000..335968ea95475ef88b3f740984435b2b13c98d34 --- /dev/null +++ b/physiological_integration/cases/case20/case20_01.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 137.645, 191.076, 217.687, 266.642, 342.243, 417.908]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.404761904762, p=0.319888641229. The response must describe the association as positive, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_02.json b/physiological_integration/cases/case20/case20_02.json new file mode 100644 index 0000000000000000000000000000000000000000..59d1103e4c355deaf9876cc7eb9fecdec17d985a --- /dev/null +++ b/physiological_integration/cases/case20/case20_02.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_03.json b/physiological_integration/cases/case20/case20_03.json new file mode 100644 index 0000000000000000000000000000000000000000..81bcbacc042a0e331281c4d5a991cf475e1e407a --- /dev/null +++ b/physiological_integration/cases/case20/case20_03.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[59.350, 83.095, 120.641, 146.429, 212.103, 279.790, 323.770, 398.157]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.6666666666666666, + "tolerance": 0.06666666666666667 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.07098765432098762, + "tolerance": 0.007098765432098763 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.666666666667, p=0.070987654321. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_04.json b/physiological_integration/cases/case20/case20_04.json new file mode 100644 index 0000000000000000000000000000000000000000..cc4b7a38a215f1c122f7122cf43cde4cf3cbdf0c --- /dev/null +++ b/physiological_integration/cases/case20/case20_04.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[59.350, 83.095, 120.641, 146.429, 212.103, 279.790, 323.770, 398.157]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.6666666666666666, + "tolerance": 0.06666666666666667 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.07098765432098762, + "tolerance": 0.007098765432098763 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.666666666667, p=0.070987654321. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_05.json b/physiological_integration/cases/case20/case20_05.json new file mode 100644 index 0000000000000000000000000000000000000000..229c287d8f45fce981c1cdb02de202aeb2fe929c --- /dev/null +++ b/physiological_integration/cases/case20/case20_05.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[59.350, 83.095, 120.641, 146.429, 212.103, 279.790, 323.770, 398.157]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.6666666666666666, + "tolerance": 0.06666666666666667 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.07098765432098762, + "tolerance": 0.007098765432098763 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.666666666667, p=0.070987654321. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_06.json b/physiological_integration/cases/case20/case20_06.json new file mode 100644 index 0000000000000000000000000000000000000000..8c72698367160f1772661e7730ee572727991d69 --- /dev/null +++ b/physiological_integration/cases/case20/case20_06.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[59.350, 83.095, 120.641, 146.429, 212.103, 279.790, 323.770, 398.157]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.6666666666666666, + "tolerance": 0.06666666666666667 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.07098765432098762, + "tolerance": 0.007098765432098763 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.666666666667, p=0.070987654321. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_07.json b/physiological_integration/cases/case20/case20_07.json new file mode 100644 index 0000000000000000000000000000000000000000..2ca88fcd775624cd68b131840465bab3d1e937e0 --- /dev/null +++ b/physiological_integration/cases/case20/case20_07.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 142.621, 201.466, 235.946, 269.111, 292.698, 368.811, 410.495]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.833333333333, p=0.0101755401235. The response must describe the association as positive, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_08.json b/physiological_integration/cases/case20/case20_08.json new file mode 100644 index 0000000000000000000000000000000000000000..f18352a455341d26f9314763cc140d42eabc6dc6 --- /dev/null +++ b/physiological_integration/cases/case20/case20_08.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 142.621, 201.466, 235.946, 269.111, 292.698, 368.811, 410.495]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.833333333333, p=0.0101755401235. The response must describe the association as positive, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_09.json b/physiological_integration/cases/case20/case20_09.json new file mode 100644 index 0000000000000000000000000000000000000000..abb72fe5476589cfae8f186bedce5f533b17d8a1 --- /dev/null +++ b/physiological_integration/cases/case20/case20_09.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 142.621, 201.466, 235.946, 269.111, 292.698, 368.811, 410.495]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.833333333333, p=0.0101755401235. The response must describe the association as positive, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_10.json b/physiological_integration/cases/case20/case20_10.json new file mode 100644 index 0000000000000000000000000000000000000000..beee6808130c816f6d1934f2ad1b2387c0583dc7 --- /dev/null +++ b/physiological_integration/cases/case20/case20_10.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[66.639, 142.621, 201.466, 235.946, 269.111, 292.698, 368.811, 410.495]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.8333333333333334, + "tolerance": 0.08333333333333334 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.010175540123456785, + "tolerance": 0.0010175540123456785 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.833333333333, p=0.0101755401235. The response must describe the association as positive, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_11.json b/physiological_integration/cases/case20/case20_11.json new file mode 100644 index 0000000000000000000000000000000000000000..e45c2c5758ddcbb84753cd9a0fb541e8f42a80a6 --- /dev/null +++ b/physiological_integration/cases/case20/case20_11.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_12.json b/physiological_integration/cases/case20/case20_12.json new file mode 100644 index 0000000000000000000000000000000000000000..861b6f1497bf32438e70907eb2085799af5459b2 --- /dev/null +++ b/physiological_integration/cases/case20/case20_12.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_13.json b/physiological_integration/cases/case20/case20_13.json new file mode 100644 index 0000000000000000000000000000000000000000..55baa47d44359e075a626936be6212fbf1f3efb9 --- /dev/null +++ b/physiological_integration/cases/case20/case20_13.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_14.json b/physiological_integration/cases/case20/case20_14.json new file mode 100644 index 0000000000000000000000000000000000000000..03fc26cf162b25cd0c81a8eba1860263b2335ffb --- /dev/null +++ b/physiological_integration/cases/case20/case20_14.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_15.json b/physiological_integration/cases/case20/case20_15.json new file mode 100644 index 0000000000000000000000000000000000000000..9e3e61a88ca0758717866bb0393f8509173893da --- /dev/null +++ b/physiological_integration/cases/case20/case20_15.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_16.json b/physiological_integration/cases/case20/case20_16.json new file mode 100644 index 0000000000000000000000000000000000000000..f21295bd4c590c512f04b845e69b94050becce96 --- /dev/null +++ b/physiological_integration/cases/case20/case20_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_17.json b/physiological_integration/cases/case20/case20_17.json new file mode 100644 index 0000000000000000000000000000000000000000..73048debb0487ce9b8e0fb54b3a24dfbd9e82ee0 --- /dev/null +++ b/physiological_integration/cases/case20/case20_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_18.json b/physiological_integration/cases/case20/case20_18.json new file mode 100644 index 0000000000000000000000000000000000000000..b68d4ef27c2e7911d6c945267e1f7bb1134cd4f3 --- /dev/null +++ b/physiological_integration/cases/case20/case20_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_19.json b/physiological_integration/cases/case20/case20_19.json new file mode 100644 index 0000000000000000000000000000000000000000..bbfc66e574631b21d76a55bb9d7e19e7a225cab1 --- /dev/null +++ b/physiological_integration/cases/case20/case20_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_20.json b/physiological_integration/cases/case20/case20_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0e755c6c82f66ce87e6d4c4e908dca6965558957 --- /dev/null +++ b/physiological_integration/cases/case20/case20_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[49.214, 119.275, 149.566, 182.722, 208.648, 350.925, 392.746, 420.655]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.8095238095238095, + "tolerance": 0.08095238095238096 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.014902667686230098, + "tolerance": 0.00149026676862301 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.809523809524, p=0.0149026676862. The response must describe the association as negative, characterize its magnitude as strong or use comparably cautious wording, and not contradict the reported coefficient. The p-value is below the conventional 0.05 threshold, so the response may describe stronger statistical evidence, while still acknowledging sampling uncertainty. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_21.json b/physiological_integration/cases/case20/case20_21.json new file mode 100644 index 0000000000000000000000000000000000000000..d901135f130c2a23321a437aa8eddfcf1c1aaf97 --- /dev/null +++ b/physiological_integration/cases/case20/case20_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 137.645, 191.076, 217.687, 266.642, 342.243, 417.908]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.404761904762, p=0.319888641229. The response must describe the association as positive, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_22.json b/physiological_integration/cases/case20/case20_22.json new file mode 100644 index 0000000000000000000000000000000000000000..862ba92194f30eeccae6a34b58c3ffaccfa0286c --- /dev/null +++ b/physiological_integration/cases/case20/case20_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 137.645, 191.076, 217.687, 266.642, 342.243, 417.908]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.404761904762, p=0.319888641229. The response must describe the association as positive, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_23.json b/physiological_integration/cases/case20/case20_23.json new file mode 100644 index 0000000000000000000000000000000000000000..8dff0cd11a484d0d42d9fd383e79ef76c758e164 --- /dev/null +++ b/physiological_integration/cases/case20/case20_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 137.645, 191.076, 217.687, 266.642, 342.243, 417.908]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.404761904762, p=0.319888641229. The response must describe the association as positive, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_24.json b/physiological_integration/cases/case20/case20_24.json new file mode 100644 index 0000000000000000000000000000000000000000..650e9a03e9ae13ca9d3ff4081c0b9d7a2a43934a --- /dev/null +++ b/physiological_integration/cases/case20/case20_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[46.554, 73.221, 137.645, 191.076, 217.687, 266.642, 342.243, 417.908]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": 0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=0.404761904762, p=0.319888641229. The response must describe the association as positive, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_25.json b/physiological_integration/cases/case20/case20_25.json new file mode 100644 index 0000000000000000000000000000000000000000..34b91a4b11a0d38c92d11d48546448f69d604057 --- /dev/null +++ b/physiological_integration/cases/case20/case20_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_26.json b/physiological_integration/cases/case20/case20_26.json new file mode 100644 index 0000000000000000000000000000000000000000..e8e73915dd2351caf27917799dfd5cf75beab7b8 --- /dev/null +++ b/physiological_integration/cases/case20/case20_26.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_27.json b/physiological_integration/cases/case20/case20_27.json new file mode 100644 index 0000000000000000000000000000000000000000..c04c5edd32f947b55268aa037abd06c66fafc3f8 --- /dev/null +++ b/physiological_integration/cases/case20/case20_27.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_28.json b/physiological_integration/cases/case20/case20_28.json new file mode 100644 index 0000000000000000000000000000000000000000..0859e3ce7f00f22a80fd19a104cc1a58300ca19e --- /dev/null +++ b/physiological_integration/cases/case20/case20_28.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_29.json b/physiological_integration/cases/case20/case20_29.json new file mode 100644 index 0000000000000000000000000000000000000000..f0d7ede9210f7fa819464e2c412c3d12a9939f8c --- /dev/null +++ b/physiological_integration/cases/case20/case20_29.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case20/case20_30.json b/physiological_integration/cases/case20/case20_30.json new file mode 100644 index 0000000000000000000000000000000000000000..1de5a4f8930192aaee04718aafa446e1cb2cbb65 --- /dev/null +++ b/physiological_integration/cases/case20/case20_30.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case20_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Assess whether event-related EEG and fNIRS responses covary across repeated task events.\n\nFrom the public event stream, use exactly these eight complete `event_code = 16` event onsets (seconds), in chronological order: `[44.143, 75.257, 117.545, 144.267, 193.312, 224.232, 286.515, 424.542]`. These event-locked epochs are pairwise non-overlapping.\n\nFor each event, estimate the baseline-corrected beta response in `[13, 30)` Hz from EEG channels `F1`, `FC1`, and `Cz`, using `[-5, 0)` seconds as baseline and `[0, 5)` seconds as the early electrophysiological response period. Use the native EEG sampling rate, a one-sided Welch PSD with a 2-second Hann window, 50% overlap, constant detrending, density scaling, and trapezoidal integration over the half-open beta band. Do not apply additional EEG filtering or resampling. Aggregate the response across the three channels to obtain one EEG value per event.\n\nFor the same events, estimate the baseline-corrected delayed HbO response from fNIRS pair `S06D06` using its paired wavelength channels and the benchmark's public fNIRS conversion procedure. Form optical density from the raw positive intensities and baseline means without an additional low-pass filter before conversion. Summarize the HbO response over `[4, 18)` seconds to obtain one hemodynamic value per event.\n\nAssess event-level EEG–HbO covariation using Spearman's rank correlation and its two-sided p-value. Briefly interpret the direction, approximate strength, and uncertainty of the association.\n\nTreat the result as cross-event evidence of electrophysiological–hemodynamic association, not as proof of causal coupling or of a mechanism established by any single event.\n\nReport only `coupling_rho`, `coupling_p_value`, and `coupling_interpretation`.", + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the following fields:\n- coupling_rho: a finite number between -1 and 1, or null.\n- coupling_p_value: a finite number between 0 and 1, or null.\n- coupling_interpretation: a brief string interpreting the direction, approximate strength, and uncertainty of the association, or null.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY \"coupling_rho\", \"coupling_p_value\", and \"coupling_interpretation\".\n5. coupling_rho must be a finite JSON number in [-1, 1], or null if missing.\n6. coupling_p_value must be a finite JSON number in [0, 1], or null if missing.\n7. coupling_interpretation must be a brief JSON string, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"coupling_rho\": , \"coupling_p_value\": , \"coupling_interpretation\": \"\"}", + "metrics": [ + { + "metric_id": "coupling_rho", + "type": "numeric_check", + "target_key": "coupling_rho", + "weight": 40, + "params": { + "gt_value": -0.40476190476190477, + "tolerance": 0.04047619047619048 + } + }, + { + "metric_id": "coupling_p_value", + "type": "numeric_check", + "target_key": "coupling_p_value", + "weight": 30, + "params": { + "gt_value": 0.3198886412288163, + "tolerance": 0.03198886412288163 + } + }, + { + "metric_id": "coupling_interpretation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge the interpretation of the repeated-event EEG--fNIRS association. The public recipe yields 8 valid paired events, Spearman rho=-0.404761904762, p=0.319888641229. The response must describe the association as negative, characterize its magnitude as moderate or use comparably cautious wording, and not contradict the reported coefficient. The p-value does not meet the conventional 0.05 threshold, so the response must acknowledge that the association is uncertain or not conventionally significant. It must treat the result as event-level evidence across repeated events, must not imply that a single event establishes neural--hemodynamic coupling, and must not make a causal claim. Output exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_01.json b/physiological_integration/cases/case21/case21_01.json new file mode 100644 index 0000000000000000000000000000000000000000..2373796964f55826b33dcc4485f2c7492e9f39ec --- /dev/null +++ b/physiological_integration/cases/case21/case21_01.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[66, 156)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.6717934502126184, + "tolerance": 0.06717934502126184 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.5497520298582302, + "tolerance": 0.05497520298582303 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_02.json b/physiological_integration/cases/case21/case21_02.json new file mode 100644 index 0000000000000000000000000000000000000000..af76465727547cd57146f838427fdcdc36952f90 --- /dev/null +++ b/physiological_integration/cases/case21/case21_02.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[94, 184)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.660369718678305, + "tolerance": 0.0660369718678305 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.5741777584765132, + "tolerance": 0.05741777584765132 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_03.json b/physiological_integration/cases/case21/case21_03.json new file mode 100644 index 0000000000000000000000000000000000000000..11d2075613d6eb3cc4bc49357e5a15766ee0d9f2 --- /dev/null +++ b/physiological_integration/cases/case21/case21_03.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[259, 349)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.7046351894642515, + "tolerance": 0.07046351894642515 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.5942743334887605, + "tolerance": 0.05942743334887605 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_04.json b/physiological_integration/cases/case21/case21_04.json new file mode 100644 index 0000000000000000000000000000000000000000..a3e6d7d6b76b3729b7ff4a48ca6737cd6a89f3ff --- /dev/null +++ b/physiological_integration/cases/case21/case21_04.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[106, 196)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.7575679319597082, + "tolerance": 0.07575679319597083 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.591153616521032, + "tolerance": 0.05911536165210321 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_05.json b/physiological_integration/cases/case21/case21_05.json new file mode 100644 index 0000000000000000000000000000000000000000..45ddba2781513926941d9863f7ef2bb843711e35 --- /dev/null +++ b/physiological_integration/cases/case21/case21_05.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[14, 104)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.30587198953419437, + "tolerance": 0.030587198953419437 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.3858910541398652, + "tolerance": 0.038589105413986524 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_06.json b/physiological_integration/cases/case21/case21_06.json new file mode 100644 index 0000000000000000000000000000000000000000..e68081ecfbd2aabdb4c856148b46af9e3c3b2683 --- /dev/null +++ b/physiological_integration/cases/case21/case21_06.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[73, 163)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.35736392655064936, + "tolerance": 0.03573639265506494 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.2526449139596451, + "tolerance": 0.02526449139596451 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_07.json b/physiological_integration/cases/case21/case21_07.json new file mode 100644 index 0000000000000000000000000000000000000000..6121f88639465ce6971aa65e9d59af213722e594 --- /dev/null +++ b/physiological_integration/cases/case21/case21_07.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[96, 186)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.45942674309929643, + "tolerance": 0.04594267430992965 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.6296687758217007, + "tolerance": 0.06296687758217007 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_08.json b/physiological_integration/cases/case21/case21_08.json new file mode 100644 index 0000000000000000000000000000000000000000..6aecfbfdfab15bf392251d4f9413122563bbac3b --- /dev/null +++ b/physiological_integration/cases/case21/case21_08.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[116, 206)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.4069235755360593, + "tolerance": 0.04069235755360593 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.6694948667536477, + "tolerance": 0.06694948667536477 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_09.json b/physiological_integration/cases/case21/case21_09.json new file mode 100644 index 0000000000000000000000000000000000000000..fdb85a45b5790c643325daa62917cd3da1dffd79 --- /dev/null +++ b/physiological_integration/cases/case21/case21_09.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[10, 100)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.3913105736048879, + "tolerance": 0.039131057360488795 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.33387901419032706, + "tolerance": 0.03338790141903271 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_10.json b/physiological_integration/cases/case21/case21_10.json new file mode 100644 index 0000000000000000000000000000000000000000..21d4046c29ffa39a4255a8e8676493c1c0c8a80e --- /dev/null +++ b/physiological_integration/cases/case21/case21_10.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[95, 185)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5783175208475924, + "tolerance": 0.05783175208475924 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.4396941362101268, + "tolerance": 0.043969413621012685 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_11.json b/physiological_integration/cases/case21/case21_11.json new file mode 100644 index 0000000000000000000000000000000000000000..eff280b0c9792805481f9c685e65902eccb3ad3a --- /dev/null +++ b/physiological_integration/cases/case21/case21_11.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[40, 130)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5735122271689171, + "tolerance": 0.05735122271689172 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.5107896599255283, + "tolerance": 0.05107896599255283 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_12.json b/physiological_integration/cases/case21/case21_12.json new file mode 100644 index 0000000000000000000000000000000000000000..9453c28a2994a5a97d3812d98bff5463271b9de7 --- /dev/null +++ b/physiological_integration/cases/case21/case21_12.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[60, 150)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.40425257916636315, + "tolerance": 0.040425257916636315 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.6152297042213637, + "tolerance": 0.06152297042213637 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_13.json b/physiological_integration/cases/case21/case21_13.json new file mode 100644 index 0000000000000000000000000000000000000000..6acb05cc276a73ec5c65ec01dadc403acb045ae4 --- /dev/null +++ b/physiological_integration/cases/case21/case21_13.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[157, 247)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.5343993010083748, + "tolerance": 0.05343993010083749 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.4651180715114367, + "tolerance": 0.04651180715114367 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_14.json b/physiological_integration/cases/case21/case21_14.json new file mode 100644 index 0000000000000000000000000000000000000000..b33f1c93f8bb57158500690bb33e16bdd1520d17 --- /dev/null +++ b/physiological_integration/cases/case21/case21_14.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[204, 294)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.4800053669544379, + "tolerance": 0.048000536695443796 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.40859593179675174, + "tolerance": 0.04085959317967518 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_15.json b/physiological_integration/cases/case21/case21_15.json new file mode 100644 index 0000000000000000000000000000000000000000..feaff6d91f77b541114c10551d449d9d9908db7a --- /dev/null +++ b/physiological_integration/cases/case21/case21_15.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[284, 374)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.6740166383238457, + "tolerance": 0.06740166383238458 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.6207461296921701, + "tolerance": 0.06207461296921701 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_16.json b/physiological_integration/cases/case21/case21_16.json new file mode 100644 index 0000000000000000000000000000000000000000..f66f325ead949b4cc245ad644ce39a7b446fa865 --- /dev/null +++ b/physiological_integration/cases/case21/case21_16.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[346, 436)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.626170564449589, + "tolerance": 0.0626170564449589 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.5602640735372916, + "tolerance": 0.05602640735372916 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_17.json b/physiological_integration/cases/case21/case21_17.json new file mode 100644 index 0000000000000000000000000000000000000000..37e9ef9048443e19bf88d38bef4235c760bbb2a2 --- /dev/null +++ b/physiological_integration/cases/case21/case21_17.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[126, 216)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.8043727725930682, + "tolerance": 0.08043727725930683 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.6442394389474426, + "tolerance": 0.06442394389474426 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_18.json b/physiological_integration/cases/case21/case21_18.json new file mode 100644 index 0000000000000000000000000000000000000000..4de9de3af5b85022bb1b684beefb91937e2efebd --- /dev/null +++ b/physiological_integration/cases/case21/case21_18.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[165, 255)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5211142915850706, + "tolerance": 0.05211142915850706 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.36000328177577345, + "tolerance": 0.03600032817757735 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_19.json b/physiological_integration/cases/case21/case21_19.json new file mode 100644 index 0000000000000000000000000000000000000000..492f224107583f852d6379414edd76fb67857c58 --- /dev/null +++ b/physiological_integration/cases/case21/case21_19.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[292, 382)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.4869019596702743, + "tolerance": 0.04869019596702743 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.3469681718028301, + "tolerance": 0.034696817180283016 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_20.json b/physiological_integration/cases/case21/case21_20.json new file mode 100644 index 0000000000000000000000000000000000000000..be063ac9cafb3f6ad29e3bc3b93bcdd68f5cc91b --- /dev/null +++ b/physiological_integration/cases/case21/case21_20.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[151, 241)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.3387890782203931, + "tolerance": 0.03387890782203931 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.6357561627524766, + "tolerance": 0.06357561627524766 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_21.json b/physiological_integration/cases/case21/case21_21.json new file mode 100644 index 0000000000000000000000000000000000000000..df8ac1f9f105179381c298d42429dacf97444ea1 --- /dev/null +++ b/physiological_integration/cases/case21/case21_21.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[130, 220)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5894899463821995, + "tolerance": 0.058948994638219954 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.5333616368385707, + "tolerance": 0.053336163683857074 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_22.json b/physiological_integration/cases/case21/case21_22.json new file mode 100644 index 0000000000000000000000000000000000000000..13b51ef97b6ac2f8e176e12df40d110c2e49ace0 --- /dev/null +++ b/physiological_integration/cases/case21/case21_22.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[343, 433)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5712469571269577, + "tolerance": 0.057124695712695774 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.642873503248702, + "tolerance": 0.0642873503248702 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_23.json b/physiological_integration/cases/case21/case21_23.json new file mode 100644 index 0000000000000000000000000000000000000000..7fb7b22d32209e5ee08afd16cb7712b956ca2689 --- /dev/null +++ b/physiological_integration/cases/case21/case21_23.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[382, 472)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.7497658825813126, + "tolerance": 0.07497658825813126 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.6523560912152032, + "tolerance": 0.06523560912152032 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_24.json b/physiological_integration/cases/case21/case21_24.json new file mode 100644 index 0000000000000000000000000000000000000000..4b14aca21d0d3fde777a81821d65ee2f6eb4f9b0 --- /dev/null +++ b/physiological_integration/cases/case21/case21_24.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[401, 491)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.5728560942716178, + "tolerance": 0.05728560942716178 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.49814693883947436, + "tolerance": 0.049814693883947436 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_25.json b/physiological_integration/cases/case21/case21_25.json new file mode 100644 index 0000000000000000000000000000000000000000..a2e48764153dcf8f1997e8fb34e823ce6cef6841 --- /dev/null +++ b/physiological_integration/cases/case21/case21_25.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[279, 369)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.2685466165800373, + "tolerance": 0.026854661658003733 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.44231468532764245, + "tolerance": 0.044231468532764245 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_26.json b/physiological_integration/cases/case21/case21_26.json new file mode 100644 index 0000000000000000000000000000000000000000..fa7d2bfc533dca6e5abb635f135367a3aa430ec5 --- /dev/null +++ b/physiological_integration/cases/case21/case21_26.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[389, 479)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.33582395725182096, + "tolerance": 0.033582395725182096 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.41923303894730374, + "tolerance": 0.041923303894730375 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_27.json b/physiological_integration/cases/case21/case21_27.json new file mode 100644 index 0000000000000000000000000000000000000000..d4c378bceeed04278c5d8f5e59c7e37e7b0bfb7c --- /dev/null +++ b/physiological_integration/cases/case21/case21_27.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[299, 389)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.3297197892493534, + "tolerance": 0.03297197892493534 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.41003865901965625, + "tolerance": 0.04100386590196563 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_28.json b/physiological_integration/cases/case21/case21_28.json new file mode 100644 index 0000000000000000000000000000000000000000..1a6a749b4b26b451c039f956858a441b6d4b91d0 --- /dev/null +++ b/physiological_integration/cases/case21/case21_28.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[344, 434)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": 0.36083197616346474, + "tolerance": 0.03608319761634648 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": 0.7132028545191116, + "tolerance": 0.07132028545191116 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_29.json b/physiological_integration/cases/case21/case21_29.json new file mode 100644 index 0000000000000000000000000000000000000000..6f27273706bc117ec3055ff36681c63f6b27df1b --- /dev/null +++ b/physiological_integration/cases/case21/case21_29.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[379, 469)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.36830280174447977, + "tolerance": 0.03683028017444798 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.4459221553801617, + "tolerance": 0.04459221553801618 + } + } + ] + } +} diff --git a/physiological_integration/cases/case21/case21_30.json b/physiological_integration/cases/case21/case21_30.json new file mode 100644 index 0000000000000000000000000000000000000000..9332341098ccede5ca007b0be9da6dd64d7bbe45 --- /dev/null +++ b/physiological_integration/cases/case21/case21_30.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case21_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze EEG–fNIRS temporal correspondence during `[399, 489)` seconds using 5-second windows stepped every 5 seconds.\n\nFor each window, summarize beta-band activity in `[13, 30)` Hz across EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`. Derive the HbO trajectory from fNIRS pair `S03D02` using the public fNIRS conversion procedure.\n\nCompute the Pearson correlation between the EEG and HbO trajectories at zero lag and with HbO delayed by 5 seconds.\n\nReport only `zero_lag_correlation` and `delayed_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG--fNIRS window-correlation task.\n\n### TASK\nExtract the two final Pearson correlations from the agent's report.\n\n### REQUIRED FIELDS\n- `zero_lag_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n- `delayed_correlation`: a finite Pearson correlation in `[-1, 1]`, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, window values, or extra fields.\n4. Do not calculate or infer a missing value.\n5. Preserve the numeric field names exactly.\n\n### OUTPUT TEMPLATE\n{\"zero_lag_correlation\": , \"delayed_correlation\": }", + "metrics": [ + { + "metric_id": "zero_lag_correlation", + "type": "numeric_check", + "target_key": "zero_lag_correlation", + "weight": 50, + "params": { + "gt_value": -0.6403796851771537, + "tolerance": 0.06403796851771537 + } + }, + { + "metric_id": "delayed_correlation", + "type": "numeric_check", + "target_key": "delayed_correlation", + "weight": 50, + "params": { + "gt_value": -0.7044742684740682, + "tolerance": 0.07044742684740683 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_01.json b/physiological_integration/cases/case22/case22_01.json new file mode 100644 index 0000000000000000000000000000000000000000..e8d38f3006877ddaaacb80ede510709acfa8ee79 --- /dev/null +++ b/physiological_integration/cases/case22/case22_01.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S01D01_760` and `S01D01_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.01995165845765727, + "tolerance": 0.0009975829228828636 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_02.json b/physiological_integration/cases/case22/case22_02.json new file mode 100644 index 0000000000000000000000000000000000000000..21ea2696ead9373ec799b5f8d08c0ca2d6f22b8b --- /dev/null +++ b/physiological_integration/cases/case22/case22_02.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S01D02_760` and `S01D02_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 2.5810532087806877, + "tolerance": 0.12905266043903438 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.019044899875732928, + "tolerance": 0.0009522449937866464 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_03.json b/physiological_integration/cases/case22/case22_03.json new file mode 100644 index 0000000000000000000000000000000000000000..a09808e2bf68c666b393b1cabd1e16ca9c9e0541 --- /dev/null +++ b/physiological_integration/cases/case22/case22_03.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S02D01_760` and `S02D01_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 2.5810532087806877, + "tolerance": 0.12905266043903438 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.02023075863418168, + "tolerance": 0.001011537931709084 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_04.json b/physiological_integration/cases/case22/case22_04.json new file mode 100644 index 0000000000000000000000000000000000000000..5a0d0c5e1d4f0bbce401c98be0f823b6ec815ad7 --- /dev/null +++ b/physiological_integration/cases/case22/case22_04.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S02D02_760` and `S02D02_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 2.5810532087806877, + "tolerance": 0.12905266043903438 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.006265466742260746, + "tolerance": 0.00031327333711303734 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_05.json b/physiological_integration/cases/case22/case22_05.json new file mode 100644 index 0000000000000000000000000000000000000000..7424539ca234f7baa1d66ce72e52425550640148 --- /dev/null +++ b/physiological_integration/cases/case22/case22_05.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S02D03_760` and `S02D03_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 2.5810532087806877, + "tolerance": 0.12905266043903438 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.01239529292934446, + "tolerance": 0.000619764646467223 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_06.json b/physiological_integration/cases/case22/case22_06.json new file mode 100644 index 0000000000000000000000000000000000000000..8fe3409e8a1ee430a256893d2e7d7ef511eb494c --- /dev/null +++ b/physiological_integration/cases/case22/case22_06.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S03D02_760` and `S03D02_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 2.5810532087806877, + "tolerance": 0.12905266043903438 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.005481471361121693, + "tolerance": 0.00027407356805608465 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_07.json b/physiological_integration/cases/case22/case22_07.json new file mode 100644 index 0000000000000000000000000000000000000000..0645b74c0d642a7ee61ff3e1069af3ff155f04be --- /dev/null +++ b/physiological_integration/cases/case22/case22_07.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S03D03_760` and `S03D03_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -10.905873880533985, + "tolerance": 0.5452936940266992 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": 0.002430819676001058, + "tolerance": 0.0001215409838000529 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_08.json b/physiological_integration/cases/case22/case22_08.json new file mode 100644 index 0000000000000000000000000000000000000000..3e5fa47a592648e4914fdb97b05a3e8466a154c5 --- /dev/null +++ b/physiological_integration/cases/case22/case22_08.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S03D04_760` and `S03D04_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -10.905873880533985, + "tolerance": 0.5452936940266992 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": 0.005346946267583733, + "tolerance": 0.00026734731337918667 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_09.json b/physiological_integration/cases/case22/case22_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e9e906903f2ef928e1dd34b6200bdb764f34be45 --- /dev/null +++ b/physiological_integration/cases/case22/case22_09.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S04D03_760` and `S04D03_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.02784947246983211, + "tolerance": 0.0013924736234916055 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_10.json b/physiological_integration/cases/case22/case22_10.json new file mode 100644 index 0000000000000000000000000000000000000000..00315c4fe7e049d99b252343170b7b8085d874b4 --- /dev/null +++ b/physiological_integration/cases/case22/case22_10.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S04D04_760` and `S04D04_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.012785657624106207, + "tolerance": 0.0006392828812053104 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_11.json b/physiological_integration/cases/case22/case22_11.json new file mode 100644 index 0000000000000000000000000000000000000000..21f21458631238f3be8548ad2cb979e5cd165940 --- /dev/null +++ b/physiological_integration/cases/case22/case22_11.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S04D09_760` and `S04D09_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.005672157981561229, + "tolerance": 0.00028360789907806147 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_12.json b/physiological_integration/cases/case22/case22_12.json new file mode 100644 index 0000000000000000000000000000000000000000..a7117c6f19955a9f442bab6dc61160b42bc959b9 --- /dev/null +++ b/physiological_integration/cases/case22/case22_12.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S05D05_760` and `S05D05_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.024646518688900512, + "tolerance": 0.0012323259344450257 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_13.json b/physiological_integration/cases/case22/case22_13.json new file mode 100644 index 0000000000000000000000000000000000000000..59f0c515a9b27317ee525b8463f3a2e8e21cd7c6 --- /dev/null +++ b/physiological_integration/cases/case22/case22_13.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S05D06_760` and `S05D06_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.042547422595987726, + "tolerance": 0.0021273711297993864 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_14.json b/physiological_integration/cases/case22/case22_14.json new file mode 100644 index 0000000000000000000000000000000000000000..9582d0b6f447c96289483a552cc40ee8b3219e6d --- /dev/null +++ b/physiological_integration/cases/case22/case22_14.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S06D05_760` and `S06D05_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": 0.7249850216535947, + "tolerance": 0.036249251082679734 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": 0.007576462056233191, + "tolerance": 0.00037882310281165955 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_15.json b/physiological_integration/cases/case22/case22_15.json new file mode 100644 index 0000000000000000000000000000000000000000..4a054bd7866267b6f8c8211ea0a223d9fdbc3937 --- /dev/null +++ b/physiological_integration/cases/case22/case22_15.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S06D06_760` and `S06D06_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -10.905873880533985, + "tolerance": 0.5452936940266992 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.008036640871586318, + "tolerance": 0.0004018320435793159 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_16.json b/physiological_integration/cases/case22/case22_16.json new file mode 100644 index 0000000000000000000000000000000000000000..85f67c41e50bdca863979580bc309e49e5199d47 --- /dev/null +++ b/physiological_integration/cases/case22/case22_16.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S07D07_760` and `S07D07_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -10.905873880533985, + "tolerance": 0.5452936940266992 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.012412098574072317, + "tolerance": 0.0006206049287036159 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_17.json b/physiological_integration/cases/case22/case22_17.json new file mode 100644 index 0000000000000000000000000000000000000000..02b1f1b5b2ace6950727006b99d4da1da5536b40 --- /dev/null +++ b/physiological_integration/cases/case22/case22_17.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S07D08_760` and `S07D08_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": 0.011812772682019267, + "tolerance": 0.0005906386341009634 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_18.json b/physiological_integration/cases/case22/case22_18.json new file mode 100644 index 0000000000000000000000000000000000000000..6a367625c2587e2db6d1e98b6411a2e443711ec9 --- /dev/null +++ b/physiological_integration/cases/case22/case22_18.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S08D07_760` and `S08D07_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.007179954864125003, + "tolerance": 0.0003589977432062502 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_19.json b/physiological_integration/cases/case22/case22_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5d5fc9b3bf6eb06e21398886c3e1a45a168cc4eb --- /dev/null +++ b/physiological_integration/cases/case22/case22_19.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S08D08_760` and `S08D08_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.009809801895540187, + "tolerance": 0.0004904900947770094 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_20.json b/physiological_integration/cases/case22/case22_20.json new file mode 100644 index 0000000000000000000000000000000000000000..d75bfdeaf3fed0eef9693c927c8ef89998e117ae --- /dev/null +++ b/physiological_integration/cases/case22/case22_20.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S09D04_760` and `S09D04_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.008404621454498343, + "tolerance": 0.00042023107272491715 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_21.json b/physiological_integration/cases/case22/case22_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ae094b60f418eed12937d73c1bfcaf7ca2a204c1 --- /dev/null +++ b/physiological_integration/cases/case22/case22_21.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S09D09_760` and `S09D09_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.004158098317580602, + "tolerance": 0.00020790491587903014 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_22.json b/physiological_integration/cases/case22/case22_22.json new file mode 100644 index 0000000000000000000000000000000000000000..f634a4518f88d7244c940a91e6c96392d490e083 --- /dev/null +++ b/physiological_integration/cases/case22/case22_22.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S09D10_760` and `S09D10_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.009605486850618449, + "tolerance": 0.00048027434253092245 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_23.json b/physiological_integration/cases/case22/case22_23.json new file mode 100644 index 0000000000000000000000000000000000000000..81864d3d709db9c56c91b6f8f6d546e5ec02d60a --- /dev/null +++ b/physiological_integration/cases/case22/case22_23.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S10D09_760` and `S10D09_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.006288953371362369, + "tolerance": 0.00031444766856811844 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_24.json b/physiological_integration/cases/case22/case22_24.json new file mode 100644 index 0000000000000000000000000000000000000000..71a5ec8cebc81404b7d6263a1c809684a862ba09 --- /dev/null +++ b/physiological_integration/cases/case22/case22_24.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S10D10_760` and `S10D10_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -6.075764251974242, + "tolerance": 0.30378821259871214 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.009550692343470581, + "tolerance": 0.0004775346171735291 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_25.json b/physiological_integration/cases/case22/case22_25.json new file mode 100644 index 0000000000000000000000000000000000000000..c4cc4de45f3739298d98c5c8cafcbcdbff06fc50 --- /dev/null +++ b/physiological_integration/cases/case22/case22_25.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S11D11_760` and `S11D11_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.002648625405324019, + "tolerance": 0.00013243127026620096 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_26.json b/physiological_integration/cases/case22/case22_26.json new file mode 100644 index 0000000000000000000000000000000000000000..378925e967c853b5fcf31ce1ec3898cf00174a18 --- /dev/null +++ b/physiological_integration/cases/case22/case22_26.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S11D12_760` and `S11D12_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.0010241062119299036, + "tolerance": 5.120531059649519e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_27.json b/physiological_integration/cases/case22/case22_27.json new file mode 100644 index 0000000000000000000000000000000000000000..7a32b698e1ec5cb7673901d1d161c546ab092dfd --- /dev/null +++ b/physiological_integration/cases/case22/case22_27.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S12D11_760` and `S12D11_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.0014764970474206711, + "tolerance": 7.382485237103357e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_28.json b/physiological_integration/cases/case22/case22_28.json new file mode 100644 index 0000000000000000000000000000000000000000..dea568edbf91a28c5f04a9a817f7bcc25f25c9a4 --- /dev/null +++ b/physiological_integration/cases/case22/case22_28.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S12D12_760` and `S12D12_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": 1.7280879973025705e-05, + "tolerance": 8.640439986512853e-07 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_29.json b/physiological_integration/cases/case22/case22_29.json new file mode 100644 index 0000000000000000000000000000000000000000..91dd731249574d5d53641be4d1537a300e8e3b42 --- /dev/null +++ b/physiological_integration/cases/case22/case22_29.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S13D13_760` and `S13D13_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.009147769457465415, + "tolerance": 0.00045738847287327074 + } + } + ] + } +} diff --git a/physiological_integration/cases/case22/case22_30.json b/physiological_integration/cases/case22/case22_30.json new file mode 100644 index 0000000000000000000000000000000000000000..caf95da541c5c74598ed298e7dda9e6b07e5b915 --- /dev/null +++ b/physiological_integration/cases/case22/case22_30.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "case22_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Compare the first half `[20, 35)` seconds with the second half `[35, 50)` seconds of the EEG and fNIRS recordings.\n\nFor EEG channels `FP1`, `AFz`, `F1`, `C3`, `Cz`, `Pz`, `O1`, and `O2`, use the physical values decoded from the EDF file and convert them from volts to microvolts. Subtract each channel's mean within each half and compute its root-mean-square amplitude in microvolts. Average the channel RMS values within each half, then report the second-half value minus the first-half value as `eeg_rms_change_uv`.\n\nFor fNIRS channels `S13D14_760` and `S13D14_850`, compute the mean positive intensity in each half. For each wavelength, calculate the relative change as `(second-half mean - first-half mean) / first-half mean`, then average the two wavelength values as `fnirs_relative_intensity_change`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract these two finite numeric results:\n1) `eeg_rms_change_uv`: the second-half minus first-half EEG RMS change.\n2) `fnirs_relative_intensity_change`: the mean relative intensity change across the two declared fNIRS wavelengths.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, units, or extra fields outside the JSON object.\n4. The JSON keys must be EXACTLY `eeg_rms_change_uv` and `fnirs_relative_intensity_change`.\n5. Both values must be finite JSON numbers, or null if the corresponding result is missing.\n6. Do not output strings in place of numbers.\n\n### OUTPUT TEMPLATE\n{\"eeg_rms_change_uv\": , \"fnirs_relative_intensity_change\": }", + "metrics": [ + { + "metric_id": "eeg_rms_change_uv", + "type": "numeric_check", + "target_key": "eeg_rms_change_uv", + "weight": 50, + "params": { + "gt_value": -23.914633755849835, + "tolerance": 1.1957316877924917 + } + }, + { + "metric_id": "fnirs_relative_intensity_change", + "type": "numeric_check", + "target_key": "fnirs_relative_intensity_change", + "weight": 50, + "params": { + "gt_value": -0.010590798756553919, + "tolerance": 0.0005295399378276959 + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_01.json b/physiological_integration/cases/case23/case23_01.json new file mode 100644 index 0000000000000000000000000000000000000000..735df11eec51011df2d92f54869d560a7245bb91 --- /dev/null +++ b/physiological_integration/cases/case23/case23_01.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 9, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_02.json b/physiological_integration/cases/case23/case23_02.json new file mode 100644 index 0000000000000000000000000000000000000000..694ebfeabda6912ebf29fc99ea86e049988104da --- /dev/null +++ b/physiological_integration/cases/case23/case23_02.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 9, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_03.json b/physiological_integration/cases/case23/case23_03.json new file mode 100644 index 0000000000000000000000000000000000000000..4d5b2565045cf1eb9b055f8fd8d795a802668001 --- /dev/null +++ b/physiological_integration/cases/case23/case23_03.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_04.json b/physiological_integration/cases/case23/case23_04.json new file mode 100644 index 0000000000000000000000000000000000000000..124de5abf4699b09a3d614a61c591c171fe1d671 --- /dev/null +++ b/physiological_integration/cases/case23/case23_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 6, + 9, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_05.json b/physiological_integration/cases/case23/case23_05.json new file mode 100644 index 0000000000000000000000000000000000000000..a99cd83feb1586518139741a8541474a5c89e607 --- /dev/null +++ b/physiological_integration/cases/case23/case23_05.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_06.json b/physiological_integration/cases/case23/case23_06.json new file mode 100644 index 0000000000000000000000000000000000000000..a379e0ae056e4ecbe7a5b8d73a9c75ef712843c8 --- /dev/null +++ b/physiological_integration/cases/case23/case23_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 9, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_07.json b/physiological_integration/cases/case23/case23_07.json new file mode 100644 index 0000000000000000000000000000000000000000..cb7cad7ef8e7368ebd17205b330ccaffdd50108c --- /dev/null +++ b/physiological_integration/cases/case23/case23_07.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_08.json b/physiological_integration/cases/case23/case23_08.json new file mode 100644 index 0000000000000000000000000000000000000000..51a07b8e135bfb177457e613dbafa605a0547a29 --- /dev/null +++ b/physiological_integration/cases/case23/case23_08.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_09.json b/physiological_integration/cases/case23/case23_09.json new file mode 100644 index 0000000000000000000000000000000000000000..fcdee42ffd68b4839407ba767a54bd2e7b6f7098 --- /dev/null +++ b/physiological_integration/cases/case23/case23_09.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 11, + 13, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_10.json b/physiological_integration/cases/case23/case23_10.json new file mode 100644 index 0000000000000000000000000000000000000000..6bf7113f48b7c3b134e623c391d6e8c793367d15 --- /dev/null +++ b/physiological_integration/cases/case23/case23_10.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 13 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_11.json b/physiological_integration/cases/case23/case23_11.json new file mode 100644 index 0000000000000000000000000000000000000000..bfa60237730a162fded9c0df31e94ffdd61163fb --- /dev/null +++ b/physiological_integration/cases/case23/case23_11.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 13 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_12.json b/physiological_integration/cases/case23/case23_12.json new file mode 100644 index 0000000000000000000000000000000000000000..f6886bf488a6b11e3b80e192052d16cd6683e55b --- /dev/null +++ b/physiological_integration/cases/case23/case23_12.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 13 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_13.json b/physiological_integration/cases/case23/case23_13.json new file mode 100644 index 0000000000000000000000000000000000000000..889b87f5c98383070dc38ade1eb4096c68c2c094 --- /dev/null +++ b/physiological_integration/cases/case23/case23_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 13, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_14.json b/physiological_integration/cases/case23/case23_14.json new file mode 100644 index 0000000000000000000000000000000000000000..021011afeff2d2e3d79494897bd59c4e664bc078 --- /dev/null +++ b/physiological_integration/cases/case23/case23_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 8, + 13 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_15.json b/physiological_integration/cases/case23/case23_15.json new file mode 100644 index 0000000000000000000000000000000000000000..af936c48fa2d31afedb093b5d2fdc432ad65f342 --- /dev/null +++ b/physiological_integration/cases/case23/case23_15.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 8, + 13, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_16.json b/physiological_integration/cases/case23/case23_16.json new file mode 100644 index 0000000000000000000000000000000000000000..324aa115e129fd36665d2993eb40a74a7947007d --- /dev/null +++ b/physiological_integration/cases/case23/case23_16.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case23_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 6, + 8, + 11, + 13, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_17.json b/physiological_integration/cases/case23/case23_17.json new file mode 100644 index 0000000000000000000000000000000000000000..bc33850ee82a3d8cfa3d726860a315d4e1dbe5cc --- /dev/null +++ b/physiological_integration/cases/case23/case23_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_18.json b/physiological_integration/cases/case23/case23_18.json new file mode 100644 index 0000000000000000000000000000000000000000..b95d03ce9b7a1e1af08abaec99ee443ac7ef7d0d --- /dev/null +++ b/physiological_integration/cases/case23/case23_18.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18, + 19 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_19.json b/physiological_integration/cases/case23/case23_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5374c22c2649bbe7bd307836e7d8c4c9a663afc4 --- /dev/null +++ b/physiological_integration/cases/case23/case23_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18, + 19 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_20.json b/physiological_integration/cases/case23/case23_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0320451e6807a7d4c7e83bf003332f89abf262b4 --- /dev/null +++ b/physiological_integration/cases/case23/case23_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18, + 19 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_21.json b/physiological_integration/cases/case23/case23_21.json new file mode 100644 index 0000000000000000000000000000000000000000..6a1a2036e96d6e8e28ffddc6bc3a50b53f6b409d --- /dev/null +++ b/physiological_integration/cases/case23/case23_21.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18, + 19 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_22.json b/physiological_integration/cases/case23/case23_22.json new file mode 100644 index 0000000000000000000000000000000000000000..9e2c36404df081aade3676348869a3a526127b84 --- /dev/null +++ b/physiological_integration/cases/case23/case23_22.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 4, + 7 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_23.json b/physiological_integration/cases/case23/case23_23.json new file mode 100644 index 0000000000000000000000000000000000000000..c8bf6981531d3f24902b3bccbd72c2bf419bd416 --- /dev/null +++ b/physiological_integration/cases/case23/case23_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 17, + 18, + 19 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_24.json b/physiological_integration/cases/case23/case23_24.json new file mode 100644 index 0000000000000000000000000000000000000000..cdc7732e6765798174340e800f726f77f10b4b98 --- /dev/null +++ b/physiological_integration/cases/case23/case23_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 1, + 4, + 7 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_25.json b/physiological_integration/cases/case23/case23_25.json new file mode 100644 index 0000000000000000000000000000000000000000..b11a47b44171dfb616f429b086438ea23b71d0d8 --- /dev/null +++ b/physiological_integration/cases/case23/case23_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_26.json b/physiological_integration/cases/case23/case23_26.json new file mode 100644 index 0000000000000000000000000000000000000000..e14ffdd7a5c419d7af6d5959efa9fa882deebfb4 --- /dev/null +++ b/physiological_integration/cases/case23/case23_26.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 3, + 8, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_27.json b/physiological_integration/cases/case23/case23_27.json new file mode 100644 index 0000000000000000000000000000000000000000..e49ebc8e4e04cf24b12ef2266bb6cab82217b5c8 --- /dev/null +++ b/physiological_integration/cases/case23/case23_27.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_28.json b/physiological_integration/cases/case23/case23_28.json new file mode 100644 index 0000000000000000000000000000000000000000..d862b3571d6a1ed3ba8d2c8f4c6372be4981e360 --- /dev/null +++ b/physiological_integration/cases/case23/case23_28.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_29.json b/physiological_integration/cases/case23/case23_29.json new file mode 100644 index 0000000000000000000000000000000000000000..3055836256f441d082e3708310427112b7aee92f --- /dev/null +++ b/physiological_integration/cases/case23/case23_29.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "case23_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 11 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_30.json b/physiological_integration/cases/case23/case23_30.json new file mode 100644 index 0000000000000000000000000000000000000000..2cc4fa914d239bdc2fdaaa6cef923c07dabbe938 --- /dev/null +++ b/physiological_integration/cases/case23/case23_30.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 11, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_31.json b/physiological_integration/cases/case23/case23_31.json new file mode 100644 index 0000000000000000000000000000000000000000..b880b2c8a3a6bf2de97f2d02319653fabb80a727 --- /dev/null +++ b/physiological_integration/cases/case23/case23_31.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 11, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_32.json b/physiological_integration/cases/case23/case23_32.json new file mode 100644 index 0000000000000000000000000000000000000000..22aeaa0fa3bc37590d81ad8f52f3ca6db3897d89 --- /dev/null +++ b/physiological_integration/cases/case23/case23_32.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 8, + 11, + 17, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_33.json b/physiological_integration/cases/case23/case23_33.json new file mode 100644 index 0000000000000000000000000000000000000000..89a80e554a8530024a8683b02fdb4876625ab467 --- /dev/null +++ b/physiological_integration/cases/case23/case23_33.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_34.json b/physiological_integration/cases/case23/case23_34.json new file mode 100644 index 0000000000000000000000000000000000000000..362e9a058a9dd6f5c9053ea82b56f49927e54e8b --- /dev/null +++ b/physiological_integration/cases/case23/case23_34.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_35.json b/physiological_integration/cases/case23/case23_35.json new file mode 100644 index 0000000000000000000000000000000000000000..04299325f890956bc13c523d3d3b025cadcad252 --- /dev/null +++ b/physiological_integration/cases/case23/case23_35.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_36.json b/physiological_integration/cases/case23/case23_36.json new file mode 100644 index 0000000000000000000000000000000000000000..c415443679c32892ff16c84a30c56d1be76a46bc --- /dev/null +++ b/physiological_integration/cases/case23/case23_36.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_37.json b/physiological_integration/cases/case23/case23_37.json new file mode 100644 index 0000000000000000000000000000000000000000..47df9924c978bea92ff36b26a57cb1c284d2e715 --- /dev/null +++ b/physiological_integration/cases/case23/case23_37.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_38.json b/physiological_integration/cases/case23/case23_38.json new file mode 100644 index 0000000000000000000000000000000000000000..5d207a5ea69004067368d3cb2fac890b114e1372 --- /dev/null +++ b/physiological_integration/cases/case23/case23_38.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_39.json b/physiological_integration/cases/case23/case23_39.json new file mode 100644 index 0000000000000000000000000000000000000000..ef3616ebcec78ea10fd6238f17bfebb705957daf --- /dev/null +++ b/physiological_integration/cases/case23/case23_39.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case23_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case23/case23_40.json b/physiological_integration/cases/case23/case23_40.json new file mode 100644 index 0000000000000000000000000000000000000000..be40df99eb04a6d81eb8e18f3f4d876b25831519 --- /dev/null +++ b/physiological_integration/cases/case23/case23_40.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case23_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Using the public trial intervals, assess cortical–pupillary arousal across the declared trials {3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18}.\n\nFor each trial, summarize frontal EEG activation as log beta power minus log alpha power across `FP1`, `FP2`, `F3`, and `F4`, using alpha `[8, 13)` Hz and beta `[13, 30)` Hz. Summarize pupil activity using mean valid pupil diameter.\n\nAcross all declared trials with valid EEG and pupil features, compute the median EEG feature and the median pupil feature separately. Select a trial only when its EEG feature is strictly greater than the EEG median and its pupil feature is strictly greater than the pupil median. Report the selected one-based trial indices as `concordant_high_arousal_trials`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract `concordant_high_arousal_trials`, the unordered set of one-based trial indices whose frontal EEG and pupil features are both above their respective medians across the valid declared trials.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, distances, rankings, or extra fields.\n3. The key must be EXACTLY `concordant_high_arousal_trials`.\n4. The value must be an array of one-based integer trial indices; order has no meaning.\n5. Extract only the final concordant set, not intermediate EEG-only or pupil-only sets.\n6. Remove duplicate trial indices.\n7. If the final set is missing, ambiguous, or not an identifiable array, return {\"concordant_high_arousal_trials\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"concordant_high_arousal_trials\": [3, 8, 12]}", + "metrics": [ + { + "metric_id": "concordant_high_arousal_trial_set", + "type": "set_match_check", + "target_key": "concordant_high_arousal_trials", + "weight": 100, + "params": { + "gt_value": [ + 7, + 11, + 14, + 18 + ], + "match_mode": "exact" + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_01.json b/physiological_integration/cases/case24/case24_01.json new file mode 100644 index 0000000000000000000000000000000000000000..6ddf5337324ab9b44f495c11cc1dc723c300b327 --- /dev/null +++ b/physiological_integration/cases/case24/case24_01.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `1` with reference trial `16` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.4039894206655509, + "tolerance": 0.04039894206655509 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 1.4075724315572682, + "tolerance": 0.14075724315572682 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.4039894206655509 and a mean bilateral pupil-diameter difference of 1.407572431557268. These values are target trial 1 minus reference trial 16.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_02.json b/physiological_integration/cases/case24/case24_02.json new file mode 100644 index 0000000000000000000000000000000000000000..a434351ffb9772f0b2817fa2abb81e69b11ad82b --- /dev/null +++ b/physiological_integration/cases/case24/case24_02.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `18` with reference trial `16` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.745504447736705, + "tolerance": 0.0745504447736705 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 1.4586763894111026, + "tolerance": 0.14586763894111027 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.745504447736705 and a mean bilateral pupil-diameter difference of 1.458676389411103. These values are target trial 18 minus reference trial 16.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_03.json b/physiological_integration/cases/case24/case24_03.json new file mode 100644 index 0000000000000000000000000000000000000000..f9dd83bea6ba710f34fe336aea661b70dc22dd22 --- /dev/null +++ b/physiological_integration/cases/case24/case24_03.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `17` with reference trial `16` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.5598818471365565, + "tolerance": 0.05598818471365566 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 1.838844932495705, + "tolerance": 0.1838844932495705 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.5598818471365565 and a mean bilateral pupil-diameter difference of 1.838844932495705. These values are target trial 17 minus reference trial 16.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_04.json b/physiological_integration/cases/case24/case24_04.json new file mode 100644 index 0000000000000000000000000000000000000000..60ed146de74e13ee7d6a7b9e37351a8bd53098cc --- /dev/null +++ b/physiological_integration/cases/case24/case24_04.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `17` with reference trial `12` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.8132559061761432, + "tolerance": 0.08132559061761432 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.7714398349565679, + "tolerance": 0.07714398349565679 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.8132559061761432 and a mean bilateral pupil-diameter difference of 0.7714398349565679. These values are target trial 17 minus reference trial 12.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_05.json b/physiological_integration/cases/case24/case24_05.json new file mode 100644 index 0000000000000000000000000000000000000000..7e2eb9eeac4c5185de62715b8f4e40a391435ff0 --- /dev/null +++ b/physiological_integration/cases/case24/case24_05.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `8` with reference trial `19` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.253506805674584, + "tolerance": 0.0253506805674584 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.7059866438137341, + "tolerance": 0.07059866438137341 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.253506805674584 and a mean bilateral pupil-diameter difference of 0.7059866438137341. These values are target trial 8 minus reference trial 19.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_06.json b/physiological_integration/cases/case24/case24_06.json new file mode 100644 index 0000000000000000000000000000000000000000..06e4a3aa66460f5c0450733a662045c094d78ec6 --- /dev/null +++ b/physiological_integration/cases/case24/case24_06.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `16` with reference trial `1` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.4039894206655509, + "tolerance": 0.04039894206655509 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.4075724315572682, + "tolerance": 0.14075724315572682 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.4039894206655509 and a mean bilateral pupil-diameter difference of -1.407572431557268. These values are target trial 16 minus reference trial 1.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_07.json b/physiological_integration/cases/case24/case24_07.json new file mode 100644 index 0000000000000000000000000000000000000000..d5d3e91efb2f174c569e2335514ea1305880595d --- /dev/null +++ b/physiological_integration/cases/case24/case24_07.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `16` with reference trial `18` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.745504447736705, + "tolerance": 0.0745504447736705 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.4586763894111026, + "tolerance": 0.14586763894111027 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.745504447736705 and a mean bilateral pupil-diameter difference of -1.458676389411103. These values are target trial 16 minus reference trial 18.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_08.json b/physiological_integration/cases/case24/case24_08.json new file mode 100644 index 0000000000000000000000000000000000000000..eedf68feda59625d6b54f698d200c613b4e8376d --- /dev/null +++ b/physiological_integration/cases/case24/case24_08.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `16` with reference trial `17` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.5598818471365565, + "tolerance": 0.05598818471365566 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.838844932495705, + "tolerance": 0.1838844932495705 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.5598818471365565 and a mean bilateral pupil-diameter difference of -1.838844932495705. These values are target trial 16 minus reference trial 17.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_09.json b/physiological_integration/cases/case24/case24_09.json new file mode 100644 index 0000000000000000000000000000000000000000..91fa6beab060ee490bf2f138bca453b22133fb6e --- /dev/null +++ b/physiological_integration/cases/case24/case24_09.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `12` with reference trial `17` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.8132559061761432, + "tolerance": 0.08132559061761432 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.7714398349565679, + "tolerance": 0.07714398349565679 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.8132559061761432 and a mean bilateral pupil-diameter difference of -0.7714398349565679. These values are target trial 12 minus reference trial 17.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_10.json b/physiological_integration/cases/case24/case24_10.json new file mode 100644 index 0000000000000000000000000000000000000000..2b5041c0158396f2689465cb06545ac430dcaf44 --- /dev/null +++ b/physiological_integration/cases/case24/case24_10.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `2` with reference trial `17` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.27768990425754103, + "tolerance": 0.027768990425754106 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.026582701522206, + "tolerance": 0.10265827015222061 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.277689904257541 and a mean bilateral pupil-diameter difference of -1.026582701522206. These values are target trial 2 minus reference trial 17.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_11.json b/physiological_integration/cases/case24/case24_11.json new file mode 100644 index 0000000000000000000000000000000000000000..079e28c96a84ca51b45e3a64890961a576164a6f --- /dev/null +++ b/physiological_integration/cases/case24/case24_11.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `15` with reference trial `8` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.25161712159471844, + "tolerance": 0.025161712159471845 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.2913756670321153, + "tolerance": 0.12913756670321153 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.2516171215947184 and a mean bilateral pupil-diameter difference of -1.291375667032115. These values are target trial 15 minus reference trial 8.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_12.json b/physiological_integration/cases/case24/case24_12.json new file mode 100644 index 0000000000000000000000000000000000000000..103b8f7a66b8add16bf37faee7573e300567fb79 --- /dev/null +++ b/physiological_integration/cases/case24/case24_12.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `4` with reference trial `7` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.4740182749309613, + "tolerance": 0.047401827493096134 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.8695775534330985, + "tolerance": 0.08695775534330985 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.4740182749309613 and a mean bilateral pupil-diameter difference of -0.8695775534330985. These values are target trial 4 minus reference trial 7.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_13.json b/physiological_integration/cases/case24/case24_13.json new file mode 100644 index 0000000000000000000000000000000000000000..6d603a1fa04898d46e7076346a45ef2d53da2aaa --- /dev/null +++ b/physiological_integration/cases/case24/case24_13.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `6` with reference trial `3` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.32867981459023043, + "tolerance": 0.032867981459023046 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.9154319585269102, + "tolerance": 0.09154319585269102 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.3286798145902304 and a mean bilateral pupil-diameter difference of -0.9154319585269102. These values are target trial 6 minus reference trial 3.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_14.json b/physiological_integration/cases/case24/case24_14.json new file mode 100644 index 0000000000000000000000000000000000000000..4f4537139c737935adde8f653761b364e470ecad --- /dev/null +++ b/physiological_integration/cases/case24/case24_14.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `5` with reference trial `14` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.3733642969245021, + "tolerance": 0.03733642969245021 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.5613671501709598, + "tolerance": 0.05613671501709599 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.3733642969245021 and a mean bilateral pupil-diameter difference of -0.5613671501709598. These values are target trial 5 minus reference trial 14.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_15.json b/physiological_integration/cases/case24/case24_15.json new file mode 100644 index 0000000000000000000000000000000000000000..ef4dfc1a44defdbd7b827b6bb74136f5227f15c4 --- /dev/null +++ b/physiological_integration/cases/case24/case24_15.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `16` with reference trial `2` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.36542472789456976, + "tolerance": 0.03654247278945698 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.8776466129112075, + "tolerance": 0.08776466129112076 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.3654247278945698 and a mean bilateral pupil-diameter difference of -0.8776466129112075. These values are target trial 16 minus reference trial 2.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_16.json b/physiological_integration/cases/case24/case24_16.json new file mode 100644 index 0000000000000000000000000000000000000000..220823e0f404db80409a866d34f1d6b85e17f64d --- /dev/null +++ b/physiological_integration/cases/case24/case24_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `17` with reference trial `15` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.2908086516306903, + "tolerance": 0.02908086516306903 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.9694426350294725, + "tolerance": 0.09694426350294726 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.2908086516306903 and a mean bilateral pupil-diameter difference of 0.9694426350294725. These values are target trial 17 minus reference trial 15.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_17.json b/physiological_integration/cases/case24/case24_17.json new file mode 100644 index 0000000000000000000000000000000000000000..1fcc672aee23f858c8817a2c0e1851da2f592b4d --- /dev/null +++ b/physiological_integration/cases/case24/case24_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `7` with reference trial `4` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.4740182749309613, + "tolerance": 0.047401827493096134 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.8695775534330985, + "tolerance": 0.08695775534330985 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.4740182749309613 and a mean bilateral pupil-diameter difference of 0.8695775534330985. These values are target trial 7 minus reference trial 4.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_18.json b/physiological_integration/cases/case24/case24_18.json new file mode 100644 index 0000000000000000000000000000000000000000..0404cd685b1dda2c481b7153cfccf80a88b48bbc --- /dev/null +++ b/physiological_integration/cases/case24/case24_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `3` with reference trial `6` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.32867981459023043, + "tolerance": 0.032867981459023046 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.9154319585269102, + "tolerance": 0.09154319585269102 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.3286798145902304 and a mean bilateral pupil-diameter difference of 0.9154319585269102. These values are target trial 3 minus reference trial 6.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_19.json b/physiological_integration/cases/case24/case24_19.json new file mode 100644 index 0000000000000000000000000000000000000000..7b33a6e8f6b9c43385ee4abcfb9acd728f84e3fa --- /dev/null +++ b/physiological_integration/cases/case24/case24_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `14` with reference trial `5` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.3733642969245021, + "tolerance": 0.03733642969245021 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.5613671501709598, + "tolerance": 0.05613671501709599 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.3733642969245021 and a mean bilateral pupil-diameter difference of 0.5613671501709598. These values are target trial 14 minus reference trial 5.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_20.json b/physiological_integration/cases/case24/case24_20.json new file mode 100644 index 0000000000000000000000000000000000000000..5dcfa5e0e697a5316482236a1936cb1fef993882 --- /dev/null +++ b/physiological_integration/cases/case24/case24_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `1` with reference trial `16` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.2632511304938525, + "tolerance": 0.026325113049385254 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.8278609006736728, + "tolerance": 0.08278609006736729 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.2632511304938525 and a mean bilateral pupil-diameter difference of 0.8278609006736728. These values are target trial 1 minus reference trial 16.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_21.json b/physiological_integration/cases/case24/case24_21.json new file mode 100644 index 0000000000000000000000000000000000000000..3f5e19b7418c410642aa2f04928193f33e70c322 --- /dev/null +++ b/physiological_integration/cases/case24/case24_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `4` with reference trial `5` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.23291602648329324, + "tolerance": 0.023291602648329326 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.8670722579246188, + "tolerance": 0.08670722579246189 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.2329160264832932 and a mean bilateral pupil-diameter difference of 0.8670722579246188. These values are target trial 4 minus reference trial 5.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_22.json b/physiological_integration/cases/case24/case24_22.json new file mode 100644 index 0000000000000000000000000000000000000000..43f8471d390c47018b1e3931e2dc8d4ec7bf5bd6 --- /dev/null +++ b/physiological_integration/cases/case24/case24_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `13` with reference trial `12` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.42709440505725266, + "tolerance": 0.04270944050572527 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.49014090115772646, + "tolerance": 0.04901409011577265 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.4270944050572527 and a mean bilateral pupil-diameter difference of 0.4901409011577265. These values are target trial 13 minus reference trial 12.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_23.json b/physiological_integration/cases/case24/case24_23.json new file mode 100644 index 0000000000000000000000000000000000000000..0575b46186c1effddd5f549a11582b0e0de33855 --- /dev/null +++ b/physiological_integration/cases/case24/case24_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `18` with reference trial `15` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.410032181381809, + "tolerance": 0.041003218138180905 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.8010075506597829, + "tolerance": 0.0801007550659783 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.410032181381809 and a mean bilateral pupil-diameter difference of 0.8010075506597829. These values are target trial 18 minus reference trial 15.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_24.json b/physiological_integration/cases/case24/case24_24.json new file mode 100644 index 0000000000000000000000000000000000000000..5a0c9e832335c45d1d158a59d55bcd4fc7d2daae --- /dev/null +++ b/physiological_integration/cases/case24/case24_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `3` with reference trial `19` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.3793869462231131, + "tolerance": 0.03793869462231131 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.7064119552664154, + "tolerance": 0.07064119552664154 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.3793869462231131 and a mean bilateral pupil-diameter difference of 0.7064119552664154. These values are target trial 3 minus reference trial 19.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_25.json b/physiological_integration/cases/case24/case24_25.json new file mode 100644 index 0000000000000000000000000000000000000000..111c9afdd97f1b845977b74287654e684708c69f --- /dev/null +++ b/physiological_integration/cases/case24/case24_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `18` with reference trial `12` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.2764775646957318, + "tolerance": 0.02764775646957318 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.5159063631462688, + "tolerance": 0.05159063631462688 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.2764775646957318 and a mean bilateral pupil-diameter difference of 0.5159063631462688. These values are target trial 18 minus reference trial 12.\n\nThe target interpretation is concordant_increase: both modalities increase and show concordant increases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_26.json b/physiological_integration/cases/case24/case24_26.json new file mode 100644 index 0000000000000000000000000000000000000000..b233fe56a91c0ed853971ba0d2100c23815be2a7 --- /dev/null +++ b/physiological_integration/cases/case24/case24_26.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `5` with reference trial `4` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.23291602648329324, + "tolerance": 0.023291602648329326 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.8670722579246188, + "tolerance": 0.08670722579246189 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.2329160264832932 and a mean bilateral pupil-diameter difference of -0.8670722579246188. These values are target trial 5 minus reference trial 4.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_27.json b/physiological_integration/cases/case24/case24_27.json new file mode 100644 index 0000000000000000000000000000000000000000..de8457bd5de0a2122c7b9951c171eed4a8252ee6 --- /dev/null +++ b/physiological_integration/cases/case24/case24_27.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `19` with reference trial `13` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.39081954647213374, + "tolerance": 0.039081954647213375 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.5227000426249298, + "tolerance": 0.05227000426249298 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.3908195464721337 and a mean bilateral pupil-diameter difference of -0.5227000426249298. These values are target trial 19 minus reference trial 13.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_28.json b/physiological_integration/cases/case24/case24_28.json new file mode 100644 index 0000000000000000000000000000000000000000..48979ad38d6c28468b7200a2667dd95f66c4b27d --- /dev/null +++ b/physiological_integration/cases/case24/case24_28.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `15` with reference trial `18` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.410032181381809, + "tolerance": 0.041003218138180905 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.8010075506597829, + "tolerance": 0.0801007550659783 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.410032181381809 and a mean bilateral pupil-diameter difference of -0.8010075506597829. These values are target trial 15 minus reference trial 18.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_29.json b/physiological_integration/cases/case24/case24_29.json new file mode 100644 index 0000000000000000000000000000000000000000..54deff0c6ad8cece9a0da65a882cbbdefd4bf1ee --- /dev/null +++ b/physiological_integration/cases/case24/case24_29.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `19` with reference trial `3` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.3793869462231131, + "tolerance": 0.03793869462231131 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.7064119552664154, + "tolerance": 0.07064119552664154 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.3793869462231131 and a mean bilateral pupil-diameter difference of -0.7064119552664154. These values are target trial 19 minus reference trial 3.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_30.json b/physiological_integration/cases/case24/case24_30.json new file mode 100644 index 0000000000000000000000000000000000000000..d9afa32e23d43ce23e152b024d871aaa263ae87c --- /dev/null +++ b/physiological_integration/cases/case24/case24_30.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `19` with reference trial `8` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.253506805674584, + "tolerance": 0.0253506805674584 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.7059866438137341, + "tolerance": 0.07059866438137341 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.253506805674584 and a mean bilateral pupil-diameter difference of -0.7059866438137341. These values are target trial 19 minus reference trial 8.\n\nThe target interpretation is concordant_decrease: both modalities decrease and show concordant decreases. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_31.json b/physiological_integration/cases/case24/case24_31.json new file mode 100644 index 0000000000000000000000000000000000000000..a6cd84ee9bd5054fa21d2660d19905f6def101a8 --- /dev/null +++ b/physiological_integration/cases/case24/case24_31.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `19` with reference trial `18` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.15627427431387866, + "tolerance": 0.015627427431387866 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.593236353762812, + "tolerance": 0.05932363537628121 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.1562742743138787 and a mean bilateral pupil-diameter difference of -0.593236353762812. These values are target trial 19 minus reference trial 18.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_32.json b/physiological_integration/cases/case24/case24_32.json new file mode 100644 index 0000000000000000000000000000000000000000..484ae6023708959c656586edeab74f761f248704 --- /dev/null +++ b/physiological_integration/cases/case24/case24_32.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `15` with reference trial `14` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.43444013750150573, + "tolerance": 0.04344401375015058 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.3755911737308857, + "tolerance": 0.03755911737308857 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.4344401375015057 and a mean bilateral pupil-diameter difference of -0.3755911737308857. These values are target trial 15 minus reference trial 14.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_33.json b/physiological_integration/cases/case24/case24_33.json new file mode 100644 index 0000000000000000000000000000000000000000..e001cf876d1c43b311affd5a3e5577a6705b88d7 --- /dev/null +++ b/physiological_integration/cases/case24/case24_33.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `5` with reference trial `20` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.2570212128375875, + "tolerance": 0.02570212128375875 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.4114789767978211, + "tolerance": 0.04114789767978211 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.2570212128375875 and a mean bilateral pupil-diameter difference of -0.4114789767978211. These values are target trial 5 minus reference trial 20.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_34.json b/physiological_integration/cases/case24/case24_34.json new file mode 100644 index 0000000000000000000000000000000000000000..a68f13508e580459b69e4d2c9bf71edc2f375d4c --- /dev/null +++ b/physiological_integration/cases/case24/case24_34.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `16` with reference trial `7` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.272110185371588, + "tolerance": 0.0272110185371588 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -1.1223168775068668, + "tolerance": 0.11223168775068669 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.272110185371588 and a mean bilateral pupil-diameter difference of -1.122316877506867. These values are target trial 16 minus reference trial 7.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_35.json b/physiological_integration/cases/case24/case24_35.json new file mode 100644 index 0000000000000000000000000000000000000000..fae42effec4363544c6dd30cec889953861d7baa --- /dev/null +++ b/physiological_integration/cases/case24/case24_35.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `5` with reference trial `13` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": 0.20089464910218405, + "tolerance": 0.020089464910218405 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": -0.6141490834542838, + "tolerance": 0.06141490834542838 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of 0.200894649102184 and a mean bilateral pupil-diameter difference of -0.6141490834542838. These values are target trial 5 minus reference trial 13.\n\nThe target interpretation is eeg_increase_pupil_decrease: EEG increases while pupil diameter decreases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_36.json b/physiological_integration/cases/case24/case24_36.json new file mode 100644 index 0000000000000000000000000000000000000000..fee69bf1d3bfc4423823be66a1788da1b1242098 --- /dev/null +++ b/physiological_integration/cases/case24/case24_36.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Compare target trial `13` with reference trial `19` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.14025708149568966, + "tolerance": 0.014025708149568966 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.6568227777654947, + "tolerance": 0.06568227777654947 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.1402570814956897 and a mean bilateral pupil-diameter difference of 0.6568227777654947. These values are target trial 13 minus reference trial 19.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_37.json b/physiological_integration/cases/case24/case24_37.json new file mode 100644 index 0000000000000000000000000000000000000000..e19b6d4415573a3225c1e03659fdaa57e88ac7a8 --- /dev/null +++ b/physiological_integration/cases/case24/case24_37.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Compare target trial `14` with reference trial `15` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.43444013750150573, + "tolerance": 0.04344401375015058 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.3755911737308857, + "tolerance": 0.03755911737308857 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.4344401375015057 and a mean bilateral pupil-diameter difference of 0.3755911737308857. These values are target trial 14 minus reference trial 15.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_38.json b/physiological_integration/cases/case24/case24_38.json new file mode 100644 index 0000000000000000000000000000000000000000..7aedda3c0127007925358bf0f39fb2418b9c0fb7 --- /dev/null +++ b/physiological_integration/cases/case24/case24_38.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Compare target trial `20` with reference trial `5` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.2570212128375875, + "tolerance": 0.02570212128375875 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.4114789767978211, + "tolerance": 0.04114789767978211 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.2570212128375875 and a mean bilateral pupil-diameter difference of 0.4114789767978211. These values are target trial 20 minus reference trial 5.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_39.json b/physiological_integration/cases/case24/case24_39.json new file mode 100644 index 0000000000000000000000000000000000000000..c61cc4ef836b93327acc04c588b1dc12ca2f4bbc --- /dev/null +++ b/physiological_integration/cases/case24/case24_39.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Compare target trial `7` with reference trial `16` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.272110185371588, + "tolerance": 0.0272110185371588 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 1.1223168775068668, + "tolerance": 0.11223168775068669 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.272110185371588 and a mean bilateral pupil-diameter difference of 1.122316877506867. These values are target trial 7 minus reference trial 16.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case24/case24_40.json b/physiological_integration/cases/case24/case24_40.json new file mode 100644 index 0000000000000000000000000000000000000000..58cf2b383a0846fa01ff1a56409150f147b0b451 --- /dev/null +++ b/physiological_integration/cases/case24/case24_40.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case24_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Compare target trial `13` with reference trial `5` using the public trial boundaries defined by each `event_code = 1` onset and its following `event_code = 2` endpoint.\n\nFor each trial and each of `FP1`, `FP2`, `AF3`, `AF4`, `F3`, and `F4`, compute alpha `[8, 14)` Hz and beta `[14, 30)` Hz power using a 2-second Hann-window Welch PSD with 50% overlap. Compute `log((beta + 1e-12) / (alpha + 1e-12))` for each channel and average across the six channels. Report target minus reference as `eeg_delta`.\n\nFor pupil diameter, compute the mean of all finite samples separately for the left and right pupil channels, then average the two eye means. Report target minus reference as `pupil_delta`.\n\nBriefly state whether EEG and pupil diameter increase or decrease in the target trial and whether their directions are concordant or discordant." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG and eye-tracking reports.\n\n### TASK\nExtract:\n- eeg_delta: target trial minus reference trial for frontal log beta-to-alpha power.\n- pupil_delta: target trial minus reference trial for mean valid bilateral pupil diameter.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_delta` and `pupil_delta`.\n4. Both values must be finite numbers or null if missing or unclear.\n5. Preserve the target-minus-reference direction.\n\n### OUTPUT TEMPLATE\n{\"eeg_delta\": , \"pupil_delta\": }", + "metrics": [ + { + "metric_id": "eeg_delta", + "type": "numeric_check", + "target_key": "eeg_delta", + "weight": 30, + "params": { + "gt_value": -0.20089464910218405, + "tolerance": 0.020089464910218405 + } + }, + { + "metric_id": "pupil_delta", + "type": "numeric_check", + "target_key": "pupil_delta", + "weight": 30, + "params": { + "gt_value": 0.6141490834542838, + "tolerance": 0.06141490834542838 + } + }, + { + "metric_id": "multimodal_direction_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge the interpretation of the target-versus-reference frontal EEG and pupil comparison.\n\nThe public analysis yields an EEG log beta-to-alpha difference of -0.200894649102184 and a mean bilateral pupil-diameter difference of 0.6141490834542838. These values are target trial 13 minus reference trial 5.\n\nThe target interpretation is eeg_decrease_pupil_increase: EEG decreases while pupil diameter increases, so the directions are discordant. A correct response must state the relative direction of both modalities and whether the changes are concordant or discordant. It must treat the result as a relative trial comparison and must not claim statistical significance, causality, a definitive emotional state, or a clinical diagnosis. Return {\"status\": true} only when all requirements are met; otherwise return {\"status\": false}. Output exactly one JSON object with the single boolean key status." + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_01.json b/physiological_integration/cases/case25/case25_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d933878852bf542cdf1c63d8fe5ab6e09f7dc6e3 --- /dev/null +++ b/physiological_integration/cases/case25/case25_01.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[304, 312)` seconds\n- `interval_B`: `[16, 24)` seconds\n- `interval_C`: `[112, 120)` seconds\n- `interval_D`: `[312, 320)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_02.json b/physiological_integration/cases/case25/case25_02.json new file mode 100644 index 0000000000000000000000000000000000000000..930c524967a604272781e825f0f63c2e27b99a7c --- /dev/null +++ b/physiological_integration/cases/case25/case25_02.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[120, 128)` seconds\n- `interval_B`: `[312, 320)` seconds\n- `interval_C`: `[304, 312)` seconds\n- `interval_D`: `[112, 120)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_03.json b/physiological_integration/cases/case25/case25_03.json new file mode 100644 index 0000000000000000000000000000000000000000..73f3df88082de0f3d66fbf68bc60e000814e1159 --- /dev/null +++ b/physiological_integration/cases/case25/case25_03.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1056, 1064)` seconds\n- `interval_B`: `[496, 504)` seconds\n- `interval_C`: `[152, 160)` seconds\n- `interval_D`: `[528, 536)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_03_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_03_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_03_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_04.json b/physiological_integration/cases/case25/case25_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4ad05710c46286adf3a71b0df1d3a48c70ed60ec --- /dev/null +++ b/physiological_integration/cases/case25/case25_04.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[128, 136)` seconds\n- `interval_B`: `[72, 80)` seconds\n- `interval_C`: `[88, 96)` seconds\n- `interval_D`: `[144, 152)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_05.json b/physiological_integration/cases/case25/case25_05.json new file mode 100644 index 0000000000000000000000000000000000000000..8895343b62553df35b25f82165cc1acd6d2353db --- /dev/null +++ b/physiological_integration/cases/case25/case25_05.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[520, 528)` seconds\n- `interval_B`: `[544, 552)` seconds\n- `interval_C`: `[496, 504)` seconds\n- `interval_D`: `[528, 536)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_05_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_05_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_05_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_06.json b/physiological_integration/cases/case25/case25_06.json new file mode 100644 index 0000000000000000000000000000000000000000..3a320d04813d0e9239604e1fa6dae07f0db8b89a --- /dev/null +++ b/physiological_integration/cases/case25/case25_06.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[312, 320)` seconds\n- `interval_B`: `[112, 120)` seconds\n- `interval_C`: `[168, 176)` seconds\n- `interval_D`: `[304, 312)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_07.json b/physiological_integration/cases/case25/case25_07.json new file mode 100644 index 0000000000000000000000000000000000000000..6d2d24fc52cba4a50d090bc4d5b2963cc7d547e7 --- /dev/null +++ b/physiological_integration/cases/case25/case25_07.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[112, 120)` seconds\n- `interval_B`: `[312, 320)` seconds\n- `interval_C`: `[304, 312)` seconds\n- `interval_D`: `[176, 184)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_08.json b/physiological_integration/cases/case25/case25_08.json new file mode 100644 index 0000000000000000000000000000000000000000..ac648a7032f80308a83db7b215517d5b222f97c3 --- /dev/null +++ b/physiological_integration/cases/case25/case25_08.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[232, 240)` seconds\n- `interval_B`: `[528, 536)` seconds\n- `interval_C`: `[1056, 1064)` seconds\n- `interval_D`: `[496, 504)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_03_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_03_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_03_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_09.json b/physiological_integration/cases/case25/case25_09.json new file mode 100644 index 0000000000000000000000000000000000000000..a22a7cf17dbfb2f3ca0e929a5d03d18f143b00e3 --- /dev/null +++ b/physiological_integration/cases/case25/case25_09.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[424, 432)` seconds\n- `interval_B`: `[432, 440)` seconds\n- `interval_C`: `[472, 480)` seconds\n- `interval_D`: `[456, 464)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_10.json b/physiological_integration/cases/case25/case25_10.json new file mode 100644 index 0000000000000000000000000000000000000000..0d438504a7c5136ab8924c0f9f90faf427721208 --- /dev/null +++ b/physiological_integration/cases/case25/case25_10.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[536, 544)` seconds\n- `interval_B`: `[496, 504)` seconds\n- `interval_C`: `[528, 536)` seconds\n- `interval_D`: `[544, 552)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_05_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_05_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_05_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_11.json b/physiological_integration/cases/case25/case25_11.json new file mode 100644 index 0000000000000000000000000000000000000000..c20dc411f79fb1743e291b555222178029cb0668 --- /dev/null +++ b/physiological_integration/cases/case25/case25_11.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[368, 376)` seconds\n- `interval_B`: `[304, 312)` seconds\n- `interval_C`: `[312, 320)` seconds\n- `interval_D`: `[288, 296)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_12.json b/physiological_integration/cases/case25/case25_12.json new file mode 100644 index 0000000000000000000000000000000000000000..05e26e0ca9258b9623112d77f00c62b9d00288f9 --- /dev/null +++ b/physiological_integration/cases/case25/case25_12.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[512, 520)` seconds\n- `interval_B`: `[520, 528)` seconds\n- `interval_C`: `[560, 568)` seconds\n- `interval_D`: `[504, 512)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_13.json b/physiological_integration/cases/case25/case25_13.json new file mode 100644 index 0000000000000000000000000000000000000000..8cfb45a00d9384d5825372049b89b1f698846c87 --- /dev/null +++ b/physiological_integration/cases/case25/case25_13.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1056, 1064)` seconds\n- `interval_B`: `[528, 536)` seconds\n- `interval_C`: `[496, 504)` seconds\n- `interval_D`: `[240, 248)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_03_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_03_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_03_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_14.json b/physiological_integration/cases/case25/case25_14.json new file mode 100644 index 0000000000000000000000000000000000000000..eeeab7a2d4d5429c01fbcc0f59ffa7de51d48212 --- /dev/null +++ b/physiological_integration/cases/case25/case25_14.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[856, 864)` seconds\n- `interval_B`: `[888, 896)` seconds\n- `interval_C`: `[912, 920)` seconds\n- `interval_D`: `[808, 816)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_15.json b/physiological_integration/cases/case25/case25_15.json new file mode 100644 index 0000000000000000000000000000000000000000..913eb640047b1dbee5accfa7ef9bc54a5a3a63a9 --- /dev/null +++ b/physiological_integration/cases/case25/case25_15.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[632, 640)` seconds\n- `interval_B`: `[528, 536)` seconds\n- `interval_C`: `[544, 552)` seconds\n- `interval_D`: `[704, 712)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_05_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_05_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_05_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_16.json b/physiological_integration/cases/case25/case25_16.json new file mode 100644 index 0000000000000000000000000000000000000000..2a3b66643df92567400af8ee69b59bca68a368a9 --- /dev/null +++ b/physiological_integration/cases/case25/case25_16.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[640, 648)` seconds\n- `interval_B`: `[712, 720)` seconds\n- `interval_C`: `[632, 640)` seconds\n- `interval_D`: `[664, 672)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_17.json b/physiological_integration/cases/case25/case25_17.json new file mode 100644 index 0000000000000000000000000000000000000000..8fd4a15ae7089be7bf19a419d6033d38692ac90a --- /dev/null +++ b/physiological_integration/cases/case25/case25_17.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[720, 728)` seconds\n- `interval_B`: `[712, 720)` seconds\n- `interval_C`: `[744, 752)` seconds\n- `interval_D`: `[736, 744)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_18.json b/physiological_integration/cases/case25/case25_18.json new file mode 100644 index 0000000000000000000000000000000000000000..160abd5b621b47006f4f9a366c3624b16dd34162 --- /dev/null +++ b/physiological_integration/cases/case25/case25_18.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[528, 536)` seconds\n- `interval_B`: `[496, 504)` seconds\n- `interval_C`: `[256, 264)` seconds\n- `interval_D`: `[1056, 1064)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_03_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_03_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_03_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_19.json b/physiological_integration/cases/case25/case25_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ff8b174f8191dc0e909c29cd750b80e21a6c990d --- /dev/null +++ b/physiological_integration/cases/case25/case25_19.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1024, 1032)` seconds\n- `interval_B`: `[1064, 1072)` seconds\n- `interval_C`: `[1040, 1048)` seconds\n- `interval_D`: `[1016, 1024)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_20.json b/physiological_integration/cases/case25/case25_20.json new file mode 100644 index 0000000000000000000000000000000000000000..a9355b2084f63bd1b08f3271e0308a616a0c5b86 --- /dev/null +++ b/physiological_integration/cases/case25/case25_20.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[544, 552)` seconds\n- `interval_B`: `[528, 536)` seconds\n- `interval_C`: `[704, 712)` seconds\n- `interval_D`: `[640, 648)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_05_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_05_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_05_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_21.json b/physiological_integration/cases/case25/case25_21.json new file mode 100644 index 0000000000000000000000000000000000000000..b71c459d586317421b9c8ac74f2c18fae7375da2 --- /dev/null +++ b/physiological_integration/cases/case25/case25_21.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[736, 744)` seconds\n- `interval_B`: `[712, 720)` seconds\n- `interval_C`: `[744, 752)` seconds\n- `interval_D`: `[728, 736)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_22.json b/physiological_integration/cases/case25/case25_22.json new file mode 100644 index 0000000000000000000000000000000000000000..fef43491bb147b0ac5a02c4a0d0791f9edfc68cf --- /dev/null +++ b/physiological_integration/cases/case25/case25_22.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[784, 792)` seconds\n- `interval_B`: `[776, 784)` seconds\n- `interval_C`: `[752, 760)` seconds\n- `interval_D`: `[768, 776)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_23.json b/physiological_integration/cases/case25/case25_23.json new file mode 100644 index 0000000000000000000000000000000000000000..6c7efbe76b135f08dffcea6a2e977ff001d6f24c --- /dev/null +++ b/physiological_integration/cases/case25/case25_23.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1056, 1064)` seconds\n- `interval_B`: `[312, 320)` seconds\n- `interval_C`: `[496, 504)` seconds\n- `interval_D`: `[528, 536)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_03_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_03_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_03_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_24.json b/physiological_integration/cases/case25/case25_24.json new file mode 100644 index 0000000000000000000000000000000000000000..5fd6497df4d5be6a34ad8da2e3a6ba3d5b47957d --- /dev/null +++ b/physiological_integration/cases/case25/case25_24.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1408, 1416)` seconds\n- `interval_B`: `[1416, 1424)` seconds\n- `interval_C`: `[1464, 1472)` seconds\n- `interval_D`: `[1496, 1504)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_25.json b/physiological_integration/cases/case25/case25_25.json new file mode 100644 index 0000000000000000000000000000000000000000..d09e7c222dac35cc27968f5f63501e58bd072bb3 --- /dev/null +++ b/physiological_integration/cases/case25/case25_25.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1696, 1704)` seconds\n- `interval_B`: `[1712, 1720)` seconds\n- `interval_C`: `[1728, 1736)` seconds\n- `interval_D`: `[1720, 1728)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_26.json b/physiological_integration/cases/case25/case25_26.json new file mode 100644 index 0000000000000000000000000000000000000000..5fcb228288a3e8eec410a9272e381785a3ced800 --- /dev/null +++ b/physiological_integration/cases/case25/case25_26.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1024, 1032)` seconds\n- `interval_B`: `[1000, 1008)` seconds\n- `interval_C`: `[1072, 1080)` seconds\n- `interval_D`: `[1040, 1048)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_27.json b/physiological_integration/cases/case25/case25_27.json new file mode 100644 index 0000000000000000000000000000000000000000..d939c811883ef26edf902411103ba649877e8bfe --- /dev/null +++ b/physiological_integration/cases/case25/case25_27.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1072, 1080)` seconds\n- `interval_B`: `[1040, 1048)` seconds\n- `interval_C`: `[1080, 1088)` seconds\n- `interval_D`: `[1104, 1112)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_28.json b/physiological_integration/cases/case25/case25_28.json new file mode 100644 index 0000000000000000000000000000000000000000..c17d00052542039fbad51a13534d72ee13ad08e4 --- /dev/null +++ b/physiological_integration/cases/case25/case25_28.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[2192, 2200)` seconds\n- `interval_B`: `[2128, 2136)` seconds\n- `interval_C`: `[2160, 2168)` seconds\n- `interval_D`: `[2168, 2176)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_29.json b/physiological_integration/cases/case25/case25_29.json new file mode 100644 index 0000000000000000000000000000000000000000..5a58cdbad4cd2c195b0a615bd89895e44950599a --- /dev/null +++ b/physiological_integration/cases/case25/case25_29.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[2232, 2240)` seconds\n- `interval_B`: `[2384, 2392)` seconds\n- `interval_C`: `[2376, 2384)` seconds\n- `interval_D`: `[2368, 2376)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_30.json b/physiological_integration/cases/case25/case25_30.json new file mode 100644 index 0000000000000000000000000000000000000000..1eed26ed85fb5e80dac7e30180c50f0c0da4dd73 --- /dev/null +++ b/physiological_integration/cases/case25/case25_30.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[2376, 2384)` seconds\n- `interval_B`: `[2616, 2624)` seconds\n- `interval_C`: `[2552, 2560)` seconds\n- `interval_D`: `[2568, 2576)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_31.json b/physiological_integration/cases/case25/case25_31.json new file mode 100644 index 0000000000000000000000000000000000000000..632f781545733584d9dc0a6558890aa06a2d3b74 --- /dev/null +++ b/physiological_integration/cases/case25/case25_31.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1104, 1112)` seconds\n- `interval_B`: `[1120, 1128)` seconds\n- `interval_C`: `[1096, 1104)` seconds\n- `interval_D`: `[1072, 1080)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_32.json b/physiological_integration/cases/case25/case25_32.json new file mode 100644 index 0000000000000000000000000000000000000000..dc7b2004932f7ae5d1ae1ad1e1e340bb465b9661 --- /dev/null +++ b/physiological_integration/cases/case25/case25_32.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1136, 1144)` seconds\n- `interval_B`: `[1120, 1128)` seconds\n- `interval_C`: `[1144, 1152)` seconds\n- `interval_D`: `[1160, 1168)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_33.json b/physiological_integration/cases/case25/case25_33.json new file mode 100644 index 0000000000000000000000000000000000000000..a4819a3a543e7ff0df53fe323e787bf48d154470 --- /dev/null +++ b/physiological_integration/cases/case25/case25_33.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[2952, 2960)` seconds\n- `interval_B`: `[2904, 2912)` seconds\n- `interval_C`: `[2888, 2896)` seconds\n- `interval_D`: `[3040, 3048)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_34.json b/physiological_integration/cases/case25/case25_34.json new file mode 100644 index 0000000000000000000000000000000000000000..4500a17cb69b08a2c3cfe68ea30910fcc4813233 --- /dev/null +++ b/physiological_integration/cases/case25/case25_34.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[3040, 3048)` seconds\n- `interval_B`: `[3184, 3192)` seconds\n- `interval_C`: `[3120, 3128)` seconds\n- `interval_D`: `[3056, 3064)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_C" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_35.json b/physiological_integration/cases/case25/case25_35.json new file mode 100644 index 0000000000000000000000000000000000000000..1534b86365acf25ee8471d0239a97b4beece038b --- /dev/null +++ b/physiological_integration/cases/case25/case25_35.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[3240, 3248)` seconds\n- `interval_B`: `[3312, 3320)` seconds\n- `interval_C`: `[3400, 3408)` seconds\n- `interval_D`: `[3256, 3264)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_36.json b/physiological_integration/cases/case25/case25_36.json new file mode 100644 index 0000000000000000000000000000000000000000..211923e3674974b3b7627a894f0b8b4d197912da --- /dev/null +++ b/physiological_integration/cases/case25/case25_36.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1144, 1152)` seconds\n- `interval_B`: `[1160, 1168)` seconds\n- `interval_C`: `[1176, 1184)` seconds\n- `interval_D`: `[1168, 1176)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_D" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_37.json b/physiological_integration/cases/case25/case25_37.json new file mode 100644 index 0000000000000000000000000000000000000000..9ef91f284e5251f2d4441df50b250267f096ef60 --- /dev/null +++ b/physiological_integration/cases/case25/case25_37.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[1208, 1216)` seconds\n- `interval_B`: `[1176, 1184)` seconds\n- `interval_C`: `[1216, 1224)` seconds\n- `interval_D`: `[1160, 1168)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_01_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_01_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_A" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_01_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_38.json b/physiological_integration/cases/case25/case25_38.json new file mode 100644 index 0000000000000000000000000000000000000000..5faf73e4549ef13c0b51d7bd659305eba5a810e4 --- /dev/null +++ b/physiological_integration/cases/case25/case25_38.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[3640, 3648)` seconds\n- `interval_B`: `[3544, 3552)` seconds\n- `interval_C`: `[3400, 3408)` seconds\n- `interval_D`: `[3424, 3432)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_39.json b/physiological_integration/cases/case25/case25_39.json new file mode 100644 index 0000000000000000000000000000000000000000..f1d634fe2ba322ab2f60dcf74ca0ff6d419bf7be --- /dev/null +++ b/physiological_integration/cases/case25/case25_39.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[4096, 4104)` seconds\n- `interval_B`: `[4040, 4048)` seconds\n- `interval_C`: `[3984, 3992)` seconds\n- `interval_D`: `[4080, 4088)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case25/case25_40.json b/physiological_integration/cases/case25/case25_40.json new file mode 100644 index 0000000000000000000000000000000000000000..56a41371e313ff97cd318aa4452f9321082b4654 --- /dev/null +++ b/physiological_integration/cases/case25/case25_40.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "case25_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "instruction": "Inspect the following four candidate intervals from the same SEED-VII EEG recording:\n\n- `interval_A`: `[4104, 4112)` seconds\n- `interval_B`: `[4336, 4344)` seconds\n- `interval_C`: `[4216, 4224)` seconds\n- `interval_D`: `[4160, 4168)` seconds\n\nUse frontal EEG channels `FP1`, `FP2`, `F3`, and `F4` together with `EOG-H` and `EOG-V`. Identify which candidate contains one clear, brief eye blink with a corresponding transient in the frontal EEG channels. The other candidates are intended to be clean comparison intervals.\n\nSelect exactly one candidate interval. Plot the selected interval with the four frontal EEG traces and both EOG traces on aligned time axes. Mark the corresponding blink-contaminated interval on the EEG traces with localized translucent red shading. Do not shade the entire plot or intervals that are not supported by the EOG blink.\n\nSave the figure as `file_check/SEEDVII_04_case25_blink_artifacts.png`.\n\nReport only the selected candidate label and the saved figure path." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological visualization reports. Your sole task is to extract the selected blink interval and the saved image path.\n\n### TASK\nExtract:\n1. the selected candidate interval label;\n2. the path of the saved EEG--EOG blink-artifact figure.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"selected_blink_interval\" and \"file_path\".\n4. selected_blink_interval must be exactly one of \"interval_A\", \"interval_B\", \"interval_C\", \"interval_D\", or null.\n5. file_path must be a string path or null.\n6. If either answer is missing or unclear, return null for that field.\n\n### OUTPUT TEMPLATE\n{\"selected_blink_interval\": \"interval_A\", \"file_path\": \"file_check/SEEDVII_04_case25_blink_artifacts.png\"}", + "metrics": [ + { + "metric_id": "blink_interval_selection", + "type": "categorical_check", + "target_key": "selected_blink_interval", + "weight": 50, + "params": { + "gt_value": "interval_B" + } + }, + { + "metric_id": "eeg_eog_blink_artifact_figure", + "type": "file_status_check", + "target_key": "file_path", + "weight": 50, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEEDVII_04_case25_blink_artifacts.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG--EOG blink-artifact visualization task. Check the image only.\n\nReturn pass only if the image clearly shows all of the following:\n1. Time-series traces from frontal EEG channels, including recognizable labels for FP1, FP2, F3, and F4 or their equivalent normalized forms.\n2. Time-series traces from both horizontal and vertical EOG channels, labeled EOG-H and EOG-V or clearly equivalent names.\n3. EEG and EOG signals displayed over the same time interval with aligned horizontal time axes.\n4. One clear, brief eye-blink event that is visibly prominent in the EOG signals.\n5. Corresponding blink-contaminated intervals visibly marked on the EEG portion of the figure using red shading, red spans, or clearly equivalent red highlighting.\n6. The red markings are localized intervals rather than the entire EEG plot being colored red.\n7. The marked EEG interval visibly corresponds in time to the prominent EOG deflection and frontal EEG transient.\n8. The figure is a physiological time-series visualization, not a PSD plot, spectrogram, topographic map, correlation matrix, bar chart, text screenshot, or unrelated image.\n9. The image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame.\n\nDo not require a particular subplot layout, line color, amplitude scale, filtering method, number of detected blinks, or exact blink timestamps. Do not judge clinical significance or whether every blink was detected. Output only one valid JSON object containing exactly the key status with value true or false: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_01.json b/physiological_integration/cases/case26/case26_01.json new file mode 100644 index 0000000000000000000000000000000000000000..a46031bb986c05d80b948c385dade4d558b18efe --- /dev/null +++ b/physiological_integration/cases/case26/case26_01.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `3` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_02.json b/physiological_integration/cases/case26/case26_02.json new file mode 100644 index 0000000000000000000000000000000000000000..8d26e3a4dda18bcc485878a0c350a90a934a9d6f --- /dev/null +++ b/physiological_integration/cases/case26/case26_02.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `3` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_03.json b/physiological_integration/cases/case26/case26_03.json new file mode 100644 index 0000000000000000000000000000000000000000..adf6b8f0a583c0bb2f3aeed21b7fa2be6345e53e --- /dev/null +++ b/physiological_integration/cases/case26/case26_03.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `2` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_04.json b/physiological_integration/cases/case26/case26_04.json new file mode 100644 index 0000000000000000000000000000000000000000..649bc9c19d3c433c91e06ffc2314a528a7e158fb --- /dev/null +++ b/physiological_integration/cases/case26/case26_04.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `2` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_05.json b/physiological_integration/cases/case26/case26_05.json new file mode 100644 index 0000000000000000000000000000000000000000..85a56aa8d304de48200c05425bb4adc7e323544f --- /dev/null +++ b/physiological_integration/cases/case26/case26_05.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `3` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_06.json b/physiological_integration/cases/case26/case26_06.json new file mode 100644 index 0000000000000000000000000000000000000000..9a19458e2a45446f58babac23ea45613f769265e --- /dev/null +++ b/physiological_integration/cases/case26/case26_06.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `5` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_07.json b/physiological_integration/cases/case26/case26_07.json new file mode 100644 index 0000000000000000000000000000000000000000..140ebf1a8e6e056c83b4ec55be3f16bef8eed9c2 --- /dev/null +++ b/physiological_integration/cases/case26/case26_07.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `5` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_08.json b/physiological_integration/cases/case26/case26_08.json new file mode 100644 index 0000000000000000000000000000000000000000..c8b6572d2220c02c315552485795bce44079db44 --- /dev/null +++ b/physiological_integration/cases/case26/case26_08.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `9` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_09.json b/physiological_integration/cases/case26/case26_09.json new file mode 100644 index 0000000000000000000000000000000000000000..f7d93321f1aa0853e6915aa2a64f49df34ea2a7e --- /dev/null +++ b/physiological_integration/cases/case26/case26_09.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `3` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_10.json b/physiological_integration/cases/case26/case26_10.json new file mode 100644 index 0000000000000000000000000000000000000000..7b1da7c807ef78fd39c93f7f5312e8fb9a2d04c0 --- /dev/null +++ b/physiological_integration/cases/case26/case26_10.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `4` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_11.json b/physiological_integration/cases/case26/case26_11.json new file mode 100644 index 0000000000000000000000000000000000000000..108ab70b6e853715f4cab5b500616174ba0e289c --- /dev/null +++ b/physiological_integration/cases/case26/case26_11.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `12` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_12.json b/physiological_integration/cases/case26/case26_12.json new file mode 100644 index 0000000000000000000000000000000000000000..055ce00a8a12d12610d51462a70bfbf9c154d32c --- /dev/null +++ b/physiological_integration/cases/case26/case26_12.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `12` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_13.json b/physiological_integration/cases/case26/case26_13.json new file mode 100644 index 0000000000000000000000000000000000000000..f5a9105e677df08b9d8de0f6f4088d7461a46c7e --- /dev/null +++ b/physiological_integration/cases/case26/case26_13.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `11` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_14.json b/physiological_integration/cases/case26/case26_14.json new file mode 100644 index 0000000000000000000000000000000000000000..cf8ecc1ef1ddc877fed3c0c3be4438379cf85ad2 --- /dev/null +++ b/physiological_integration/cases/case26/case26_14.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `4` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_15.json b/physiological_integration/cases/case26/case26_15.json new file mode 100644 index 0000000000000000000000000000000000000000..446b605d7dd90f5ff0656c2d263f02866346467e --- /dev/null +++ b/physiological_integration/cases/case26/case26_15.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `7` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_16.json b/physiological_integration/cases/case26/case26_16.json new file mode 100644 index 0000000000000000000000000000000000000000..4f5240770cf372d7f3f7e96e9e23eb816e78abdd --- /dev/null +++ b/physiological_integration/cases/case26/case26_16.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `14` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_17.json b/physiological_integration/cases/case26/case26_17.json new file mode 100644 index 0000000000000000000000000000000000000000..9363a85000e2f015b7ee7e536410c1ad6ccdabf4 --- /dev/null +++ b/physiological_integration/cases/case26/case26_17.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `14` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_18.json b/physiological_integration/cases/case26/case26_18.json new file mode 100644 index 0000000000000000000000000000000000000000..fd52ff966dcebf9ce8692b771453adefd9e5d52f --- /dev/null +++ b/physiological_integration/cases/case26/case26_18.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `12` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_19.json b/physiological_integration/cases/case26/case26_19.json new file mode 100644 index 0000000000000000000000000000000000000000..e46ae4ead42b19e233e0815d8240d51a15c0464e --- /dev/null +++ b/physiological_integration/cases/case26/case26_19.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `10` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_20.json b/physiological_integration/cases/case26/case26_20.json new file mode 100644 index 0000000000000000000000000000000000000000..ada9460bba54a006cbef45cd8883bd0ea3029489 --- /dev/null +++ b/physiological_integration/cases/case26/case26_20.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `12` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_21.json b/physiological_integration/cases/case26/case26_21.json new file mode 100644 index 0000000000000000000000000000000000000000..d191036e0cac74bfa2a43688c4c86634bdcc823e --- /dev/null +++ b/physiological_integration/cases/case26/case26_21.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `16` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_22.json b/physiological_integration/cases/case26/case26_22.json new file mode 100644 index 0000000000000000000000000000000000000000..dbde4ae30550faf5754622fe5f086d6827f0a023 --- /dev/null +++ b/physiological_integration/cases/case26/case26_22.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `15` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_23.json b/physiological_integration/cases/case26/case26_23.json new file mode 100644 index 0000000000000000000000000000000000000000..5782d7e80697cafbe85adda7d95df62cb31aec1b --- /dev/null +++ b/physiological_integration/cases/case26/case26_23.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `15` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_24.json b/physiological_integration/cases/case26/case26_24.json new file mode 100644 index 0000000000000000000000000000000000000000..dbbcf95241bede03f6ea0f4af6fa41b477fd641f --- /dev/null +++ b/physiological_integration/cases/case26/case26_24.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `13` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_25.json b/physiological_integration/cases/case26/case26_25.json new file mode 100644 index 0000000000000000000000000000000000000000..9550830636a0f92adff124e815291e5ce16def66 --- /dev/null +++ b/physiological_integration/cases/case26/case26_25.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `14` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_26.json b/physiological_integration/cases/case26/case26_26.json new file mode 100644 index 0000000000000000000000000000000000000000..7a7a6c94d77af38aaf192731a7eb74c3777366b0 --- /dev/null +++ b/physiological_integration/cases/case26/case26_26.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `17` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_27.json b/physiological_integration/cases/case26/case26_27.json new file mode 100644 index 0000000000000000000000000000000000000000..e263f6626ec20ccb0d430f40eba691b4b775ae56 --- /dev/null +++ b/physiological_integration/cases/case26/case26_27.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `16` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_28.json b/physiological_integration/cases/case26/case26_28.json new file mode 100644 index 0000000000000000000000000000000000000000..8364b61fd111a32893829b8d1d9e410d7b95b636 --- /dev/null +++ b/physiological_integration/cases/case26/case26_28.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `16` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_29.json b/physiological_integration/cases/case26/case26_29.json new file mode 100644 index 0000000000000000000000000000000000000000..ab3a266b96a8e25e1092711bd80426ec283b30d5 --- /dev/null +++ b/physiological_integration/cases/case26/case26_29.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `14` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_30.json b/physiological_integration/cases/case26/case26_30.json new file mode 100644 index 0000000000000000000000000000000000000000..b7cfcba50684f0078ae1b5236593dc40566e5919 --- /dev/null +++ b/physiological_integration/cases/case26/case26_30.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `16` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_31.json b/physiological_integration/cases/case26/case26_31.json new file mode 100644 index 0000000000000000000000000000000000000000..528e6da3112b2df44c99f275f283b04aad295c8c --- /dev/null +++ b/physiological_integration/cases/case26/case26_31.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `18` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_32.json b/physiological_integration/cases/case26/case26_32.json new file mode 100644 index 0000000000000000000000000000000000000000..33c44b649c8781594e6a8e9e7f316799658b8611 --- /dev/null +++ b/physiological_integration/cases/case26/case26_32.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `17` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_33.json b/physiological_integration/cases/case26/case26_33.json new file mode 100644 index 0000000000000000000000000000000000000000..542f21299d0d44bcfd849b8f3fd73b7922c99fa1 --- /dev/null +++ b/physiological_integration/cases/case26/case26_33.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `18` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_34.json b/physiological_integration/cases/case26/case26_34.json new file mode 100644 index 0000000000000000000000000000000000000000..28e08b27b82ab764d2c1c1ca760158dda5feb10c --- /dev/null +++ b/physiological_integration/cases/case26/case26_34.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `18` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_35.json b/physiological_integration/cases/case26/case26_35.json new file mode 100644 index 0000000000000000000000000000000000000000..d47814f4bc0f093fc499cb624fbd5298776c2bd3 --- /dev/null +++ b/physiological_integration/cases/case26/case26_35.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `19` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_36.json b/physiological_integration/cases/case26/case26_36.json new file mode 100644 index 0000000000000000000000000000000000000000..32c44ec1b1478c2c5d2e5f0c2e3bf81a1db0ade4 --- /dev/null +++ b/physiological_integration/cases/case26/case26_36.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze public trial `19` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_37.json b/physiological_integration/cases/case26/case26_37.json new file mode 100644 index 0000000000000000000000000000000000000000..bb5966391369478125d8257b71af58eba788965a --- /dev/null +++ b/physiological_integration/cases/case26/case26_37.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze public trial `18` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_38.json b/physiological_integration/cases/case26/case26_38.json new file mode 100644 index 0000000000000000000000000000000000000000..fe464204116904210e608e35c0ac9cb040a88d90 --- /dev/null +++ b/physiological_integration/cases/case26/case26_38.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze public trial `19` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_39.json b/physiological_integration/cases/case26/case26_39.json new file mode 100644 index 0000000000000000000000000000000000000000..2b4e4e009ca6a5971ca753c15f5e1f6020d4da40 --- /dev/null +++ b/physiological_integration/cases/case26/case26_39.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze public trial `19` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case26/case26_40.json b/physiological_integration/cases/case26/case26_40.json new file mode 100644 index 0000000000000000000000000000000000000000..27313edebadfc96a1dc069e85003e3c0cf370813 --- /dev/null +++ b/physiological_integration/cases/case26/case26_40.json @@ -0,0 +1,30 @@ +{ + "meta_info": { + "case_id": "case26_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze public trial `20` after aligning the EEG and eye-tracking recordings.\n\nDivide the trial into consecutive 2-second epochs. For each epoch, compute mean alpha power in [8, 13) Hz across P3, P4, PO3, PO4, O1, and O2. Treat a gaze sample as valid when both gaze coordinates are finite and at least one of `ValidLeft` or `ValidRight` is positive; do not interpolate missing samples. Use only epochs with finite posterior alpha power and at least half valid gaze samples.\n\nSelect the lowest quartile of valid epochs by posterior alpha power. Pool the valid gaze samples from these epochs and determine whether at least 50% fall within the central 30% × 30% region of the screen.\n\nBriefly report whether low-posterior-alpha epochs show central gaze concentration. Treat the result as a descriptive within-trial EEG–gaze association, not as evidence of causality or a specific cognitive state." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for an EEG–gaze analysis.\n\n### TASK\nExtract `central_gaze_concentration` from the agent's final answer.\n\nSet it to:\n- true when the agent reports that at least 50% of valid gaze samples from the selected low-posterior-alpha epochs fall within the central 30% × 30% screen region;\n- false when the agent reports that the proportion is below 50%;\n- null when the conclusion is missing or unclear.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The only key must be `central_gaze_concentration`.\n3. The value must be the JSON boolean true, the JSON boolean false, or null.\n4. Do not use the strings \"true\" or \"false\".\n5. Do not output Markdown, explanations, numerical values, or extra fields.\n\n### OUTPUT TEMPLATE\n{\"central_gaze_concentration\": }", + "metrics": [ + { + "metric_id": "central_gaze_concentration", + "type": "categorical_check", + "target_key": "central_gaze_concentration", + "weight": 100, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_01.json b/physiological_integration/cases/case27/case27_01.json new file mode 100644 index 0000000000000000000000000000000000000000..3bfcb3ca3e1e4ca1936440864e8ecf126cb107e0 --- /dev/null +++ b/physiological_integration/cases/case27/case27_01.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.936005907796181, + "tolerance": 0.0936005907796181 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.936005907796, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_02.json b/physiological_integration/cases/case27/case27_02.json new file mode 100644 index 0000000000000000000000000000000000000000..2f8e980dcbf9504d695d0a2df85c644ee861f0a7 --- /dev/null +++ b/physiological_integration/cases/case27/case27_02.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.9013187045046945, + "tolerance": 0.09013187045046946 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the frontal ROI and central_gaze_fraction=0.901318704505, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_03.json b/physiological_integration/cases/case27/case27_03.json new file mode 100644 index 0000000000000000000000000000000000000000..205c4c62dc540bcfe48ad9b780b110553f0db78f --- /dev/null +++ b/physiological_integration/cases/case27/case27_03.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.92771236865426, + "tolerance": 0.092771236865426 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.927712368654, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_04.json b/physiological_integration/cases/case27/case27_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4b52001d5f39a8455ccd82a3fee086e661b92bcd --- /dev/null +++ b/physiological_integration/cases/case27/case27_04.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the first chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7865255549819308, + "tolerance": 0.07865255549819308 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.786525554982, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_05.json b/physiological_integration/cases/case27/case27_05.json new file mode 100644 index 0000000000000000000000000000000000000000..78b0024f756ecb079746d1dfe362da5c89ab4d6c --- /dev/null +++ b/physiological_integration/cases/case27/case27_05.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the second chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.677794191828205, + "tolerance": 0.0677794191828205 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.677794191828, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_06.json b/physiological_integration/cases/case27/case27_06.json new file mode 100644 index 0000000000000000000000000000000000000000..d93f7c3237d84d4e77c0e7af1be0a2e08b3fdcbc --- /dev/null +++ b/physiological_integration/cases/case27/case27_06.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the fifth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.8465249804605303, + "tolerance": 0.08465249804605303 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.846524980461, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_07.json b/physiological_integration/cases/case27/case27_07.json new file mode 100644 index 0000000000000000000000000000000000000000..030e1063849798292e77c68dc2b67dfd291b9997 --- /dev/null +++ b/physiological_integration/cases/case27/case27_07.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the tenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7139466424178906, + "tolerance": 0.07139466424178907 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.713946642418, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_08.json b/physiological_integration/cases/case27/case27_08.json new file mode 100644 index 0000000000000000000000000000000000000000..61d696143392ceb36e39d33b5957d48ec76150b0 --- /dev/null +++ b/physiological_integration/cases/case27/case27_08.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the eighth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.6442758253461129, + "tolerance": 0.0644275825346113 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.644275825346, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_09.json b/physiological_integration/cases/case27/case27_09.json new file mode 100644 index 0000000000000000000000000000000000000000..f26677baa1184af7c2840fb9d16e9cb859504b44 --- /dev/null +++ b/physiological_integration/cases/case27/case27_09.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the ninth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7385049222404215, + "tolerance": 0.07385049222404215 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.73850492224, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_10.json b/physiological_integration/cases/case27/case27_10.json new file mode 100644 index 0000000000000000000000000000000000000000..cc0a7dc5aa08092c107e6d26971149dc17624930 --- /dev/null +++ b/physiological_integration/cases/case27/case27_10.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.9313626197408955, + "tolerance": 0.09313626197408956 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.931362619741, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_11.json b/physiological_integration/cases/case27/case27_11.json new file mode 100644 index 0000000000000000000000000000000000000000..cb69b7f763e3afb26d7170808c0d2cc6a92338e9 --- /dev/null +++ b/physiological_integration/cases/case27/case27_11.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the eighth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.8052767498202923, + "tolerance": 0.08052767498202923 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.80527674982, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_12.json b/physiological_integration/cases/case27/case27_12.json new file mode 100644 index 0000000000000000000000000000000000000000..08acc17f2c375ee1341d72742f5acca20c2beeda --- /dev/null +++ b/physiological_integration/cases/case27/case27_12.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the twelfth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.6716895224673424, + "tolerance": 0.06716895224673423 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.671689522467, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_13.json b/physiological_integration/cases/case27/case27_13.json new file mode 100644 index 0000000000000000000000000000000000000000..0226a5694fbdaa7c3274eb9985a58ab7049acd82 --- /dev/null +++ b/physiological_integration/cases/case27/case27_13.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the first chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.5009038089089736, + "tolerance": 0.05009038089089736 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the frontal ROI and central_gaze_fraction=0.500903808909, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_14.json b/physiological_integration/cases/case27/case27_14.json new file mode 100644 index 0000000000000000000000000000000000000000..caf077d63be464509ce96dadfc0b6029082de960 --- /dev/null +++ b/physiological_integration/cases/case27/case27_14.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the fourteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.8888159813356664, + "tolerance": 0.08888159813356665 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.888815981336, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_15.json b/physiological_integration/cases/case27/case27_15.json new file mode 100644 index 0000000000000000000000000000000000000000..30f9d389575007b24d54ad1635413ac9214aa4ca --- /dev/null +++ b/physiological_integration/cases/case27/case27_15.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the fourth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.759204776909284, + "tolerance": 0.07592047769092841 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the frontal ROI and central_gaze_fraction=0.759204776909, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_16.json b/physiological_integration/cases/case27/case27_16.json new file mode 100644 index 0000000000000000000000000000000000000000..b9cb5d8466272959f4c97b203a353435b1528e02 --- /dev/null +++ b/physiological_integration/cases/case27/case27_16.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the ninth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.6564790111782783, + "tolerance": 0.06564790111782784 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.656479011178, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_17.json b/physiological_integration/cases/case27/case27_17.json new file mode 100644 index 0000000000000000000000000000000000000000..d40dec6a84236348ec7c4bff07dc82abd0af044d --- /dev/null +++ b/physiological_integration/cases/case27/case27_17.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the thirteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.6400364289508969, + "tolerance": 0.06400364289508968 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.640036428951, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_18.json b/physiological_integration/cases/case27/case27_18.json new file mode 100644 index 0000000000000000000000000000000000000000..9dc85eeeda6348614d06a5f5969398b6a48d16fa --- /dev/null +++ b/physiological_integration/cases/case27/case27_18.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the eighteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.693833780160858, + "tolerance": 0.0693833780160858 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.693833780161, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_19.json b/physiological_integration/cases/case27/case27_19.json new file mode 100644 index 0000000000000000000000000000000000000000..625d16f2a5e16c4349c72d2cfc03d39e7dc30915 --- /dev/null +++ b/physiological_integration/cases/case27/case27_19.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.928915052538296, + "tolerance": 0.09289150525382961 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.928915052538, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_20.json b/physiological_integration/cases/case27/case27_20.json new file mode 100644 index 0000000000000000000000000000000000000000..b197efcd9bb024799b648898612920a3549a091c --- /dev/null +++ b/physiological_integration/cases/case27/case27_20.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the sixth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7041971168741229, + "tolerance": 0.07041971168741229 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.704197116874, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_21.json b/physiological_integration/cases/case27/case27_21.json new file mode 100644 index 0000000000000000000000000000000000000000..a17b85a03eaff1b8197ad988865df6c867f27dfd --- /dev/null +++ b/physiological_integration/cases/case27/case27_21.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the fourteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.945647419072616, + "tolerance": 0.0945647419072616 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.945647419073, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_22.json b/physiological_integration/cases/case27/case27_22.json new file mode 100644 index 0000000000000000000000000000000000000000..ae990288140aa7ee6d8700a23d2357533681348b --- /dev/null +++ b/physiological_integration/cases/case27/case27_22.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the fourteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7676927843694753, + "tolerance": 0.07676927843694753 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.767692784369, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_23.json b/physiological_integration/cases/case27/case27_23.json new file mode 100644 index 0000000000000000000000000000000000000000..ece6718198fcc07b2494e20ab4206d49130cb29d --- /dev/null +++ b/physiological_integration/cases/case27/case27_23.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the second chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.40539371993730744, + "tolerance": 0.04053937199373075 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.405393719937, which is below 0.5, indicating relatively peripheral or not centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_24.json b/physiological_integration/cases/case27/case27_24.json new file mode 100644 index 0000000000000000000000000000000000000000..be669e6e611f506d808e3a65ec0db9b8598c754c --- /dev/null +++ b/physiological_integration/cases/case27/case27_24.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the sixth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7372430383437819, + "tolerance": 0.0737243038343782 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.737243038344, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_25.json b/physiological_integration/cases/case27/case27_25.json new file mode 100644 index 0000000000000000000000000000000000000000..4c33981a32cc74d8085abf36183dc3ef4d1fbeb1 --- /dev/null +++ b/physiological_integration/cases/case27/case27_25.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the eighth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.8068423855165069, + "tolerance": 0.08068423855165069 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.806842385517, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_26.json b/physiological_integration/cases/case27/case27_26.json new file mode 100644 index 0000000000000000000000000000000000000000..b16e69a1403e3265fd6708ba7916d809e0404c22 --- /dev/null +++ b/physiological_integration/cases/case27/case27_26.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "Analyze the third chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.936005907796181, + "tolerance": 0.0936005907796181 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.936005907796, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_27.json b/physiological_integration/cases/case27/case27_27.json new file mode 100644 index 0000000000000000000000000000000000000000..f689a470b6ae6555fe592455e152a2456e619d1c --- /dev/null +++ b/physiological_integration/cases/case27/case27_27.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "Analyze the fifteenth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.585722422631479, + "tolerance": 0.0585722422631479 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.585722422631, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_28.json b/physiological_integration/cases/case27/case27_28.json new file mode 100644 index 0000000000000000000000000000000000000000..5d0ea764938bdf11b8f04535d2f1bc1d1bb9250d --- /dev/null +++ b/physiological_integration/cases/case27/case27_28.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "Analyze the sixth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.6907669315447984, + "tolerance": 0.06907669315447984 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.690766931545, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_29.json b/physiological_integration/cases/case27/case27_29.json new file mode 100644 index 0000000000000000000000000000000000000000..ffca6e140e0ebf877cf015a9421e4f9430b8e32a --- /dev/null +++ b/physiological_integration/cases/case27/case27_29.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "Analyze the seventh chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.7172619946254416, + "tolerance": 0.07172619946254416 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.717261994625, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case27/case27_30.json b/physiological_integration/cases/case27/case27_30.json new file mode 100644 index 0000000000000000000000000000000000000000..2bbcb76ce2fef2aec01eb937885646a29f4bd1e3 --- /dev/null +++ b/physiological_integration/cases/case27/case27_30.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "case27_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "Analyze the ninth chronologically ordered event_code=1 → event_code=2 interval.\nUse the following EEG ROIs:\n{\n \"frontal\": [\"FP1\", \"FP2\", \"F3\", \"F4\"],\n \"posterior\": [\"P3\", \"P4\", \"O1\", \"O2\"]\n}\nCompare mean alpha power in [8, 13) Hz between the two ROIs. Report the ROI with lower alpha power as lower_alpha_roi.\nReport the proportion of finite gaze samples in the central 50% × 50% screen region as central_gaze_fraction.\nBriefly state which ROI has lower alpha power, whether gaze is centrally concentrated or relatively peripheral using the 0.5 threshold, and describe the two findings together as a descriptive EEG–gaze spatial profile. Do not interpret lower alpha as a direct measure of engagement or imply causal EEG–gaze coupling.\n\nReport only lower_alpha_roi and central_gaze_fraction, together with the brief descriptive interpretation." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG and eye-tracking reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract lower_alpha_roi (frontal or posterior) and central_gaze_fraction (the proportion of finite gaze samples in the central screen region).\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY lower_alpha_roi and central_gaze_fraction.\n4. lower_alpha_roi must be exactly frontal, posterior, or null if missing or unclear.\n5. central_gaze_fraction must be a finite number in [0, 1], or null if missing.\n6. If a percentage is reported, convert it to a proportion between 0 and 1.\n7. Do not extract a combined profile label as a separate field.\n\n### OUTPUT TEMPLATE\n{\"lower_alpha_roi\": \"\", \"central_gaze_fraction\": }", + "metrics": [ + { + "metric_id": "lower_alpha_roi", + "type": "categorical_check", + "target_key": "lower_alpha_roi", + "weight": 35, + "params": { + "gt_value": "posterior" + } + }, + { + "metric_id": "central_gaze_fraction", + "type": "numeric_check", + "target_key": "central_gaze_fraction", + "weight": 35, + "params": { + "gt_value": 0.5628313802303199, + "tolerance": 0.05628313802303199 + } + }, + { + "metric_id": "eeg_gaze_profile_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge only the natural-language interpretation of the regional EEG alpha and gaze-distribution results. Do not score the exact structured fields; those are evaluated separately.\n\nThe public result is lower alpha in the posterior ROI and central_gaze_fraction=0.56283138023, which is at least 0.5, indicating centrally concentrated gaze.\n\nA correct response must state which ROI has lower alpha, describe gaze as central or relatively peripheral according to 0.5, and combine the two findings into a descriptive EEG--gaze spatial profile. It must not reverse either finding, treat lower alpha as a direct measure of engagement, claim causal EEG--gaze coupling, claim statistical significance, infer emotional or clinical state, or make a clinical diagnosis. Exact wording is not required.\n\nOutput exactly one JSON object: {\"status\": true} or {\"status\": false}. Do not output Markdown, explanations, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_01.json b/physiological_integration/cases/case28/case28_01.json new file mode 100644 index 0000000000000000000000000000000000000000..ea246520cade967766613c6a1381976ee041952c --- /dev/null +++ b/physiological_integration/cases/case28/case28_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `31`, compare the 3-second baseline `[1890, 1893)` with the stimulus period `[1893, 1953)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_02.json b/physiological_integration/cases/case28/case28_02.json new file mode 100644 index 0000000000000000000000000000000000000000..dbaf202223e549c471a24d43549f3ca25b1d1f68 --- /dev/null +++ b/physiological_integration/cases/case28/case28_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `18`, compare the 3-second baseline `[1071, 1074)` with the stimulus period `[1074, 1134)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_03.json b/physiological_integration/cases/case28/case28_03.json new file mode 100644 index 0000000000000000000000000000000000000000..348568c52cc1b7253da1305a5454d49bbc1d234e --- /dev/null +++ b/physiological_integration/cases/case28/case28_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `33`, compare the 3-second baseline `[2016, 2019)` with the stimulus period `[2019, 2079)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_04.json b/physiological_integration/cases/case28/case28_04.json new file mode 100644 index 0000000000000000000000000000000000000000..28904fed88174f77ee7499cc19c480c1122e4943 --- /dev/null +++ b/physiological_integration/cases/case28/case28_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `16`, compare the 3-second baseline `[945, 948)` with the stimulus period `[948, 1008)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_05.json b/physiological_integration/cases/case28/case28_05.json new file mode 100644 index 0000000000000000000000000000000000000000..17ea20926e684f18c0173d2066d5445ef6fcbc59 --- /dev/null +++ b/physiological_integration/cases/case28/case28_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `39`, compare the 3-second baseline `[2394, 2397)` with the stimulus period `[2397, 2457)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_06.json b/physiological_integration/cases/case28/case28_06.json new file mode 100644 index 0000000000000000000000000000000000000000..614fd091a7d5d0dc2d96c937cba6f61df121dda9 --- /dev/null +++ b/physiological_integration/cases/case28/case28_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `35`, compare the 3-second baseline `[2142, 2145)` with the stimulus period `[2145, 2205)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_07.json b/physiological_integration/cases/case28/case28_07.json new file mode 100644 index 0000000000000000000000000000000000000000..2815fda8899b5ee644eeeec4f656cc639dfeeafa --- /dev/null +++ b/physiological_integration/cases/case28/case28_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `15`, compare the 3-second baseline `[882, 885)` with the stimulus period `[885, 945)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_08.json b/physiological_integration/cases/case28/case28_08.json new file mode 100644 index 0000000000000000000000000000000000000000..b902b43f2a2ebdcfead4cfd49f7a7f7b2fdeddca --- /dev/null +++ b/physiological_integration/cases/case28/case28_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `28`, compare the 3-second baseline `[1701, 1704)` with the stimulus period `[1704, 1764)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_09.json b/physiological_integration/cases/case28/case28_09.json new file mode 100644 index 0000000000000000000000000000000000000000..68279e99780c9479fc5178ffc488d0304f689a44 --- /dev/null +++ b/physiological_integration/cases/case28/case28_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `36`, compare the 3-second baseline `[2205, 2208)` with the stimulus period `[2208, 2268)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_10.json b/physiological_integration/cases/case28/case28_10.json new file mode 100644 index 0000000000000000000000000000000000000000..2961efa64612ffa0f2caea54f6a84926d5c6e79d --- /dev/null +++ b/physiological_integration/cases/case28/case28_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `33`, compare the 3-second baseline `[2016, 2019)` with the stimulus period `[2019, 2079)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_11.json b/physiological_integration/cases/case28/case28_11.json new file mode 100644 index 0000000000000000000000000000000000000000..dd27f8abee65dfe5bc12724859c46f7c39e75003 --- /dev/null +++ b/physiological_integration/cases/case28/case28_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `31`, compare the 3-second baseline `[1890, 1893)` with the stimulus period `[1893, 1953)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_12.json b/physiological_integration/cases/case28/case28_12.json new file mode 100644 index 0000000000000000000000000000000000000000..a50dcbf75210ae6e8dba526a9a58afb57d191cb4 --- /dev/null +++ b/physiological_integration/cases/case28/case28_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `25`, compare the 3-second baseline `[1512, 1515)` with the stimulus period `[1515, 1575)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_13.json b/physiological_integration/cases/case28/case28_13.json new file mode 100644 index 0000000000000000000000000000000000000000..3a1020234769a796a830ced900d389b71e4d0886 --- /dev/null +++ b/physiological_integration/cases/case28/case28_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `7`, compare the 3-second baseline `[378, 381)` with the stimulus period `[381, 441)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_14.json b/physiological_integration/cases/case28/case28_14.json new file mode 100644 index 0000000000000000000000000000000000000000..7d1247031d8572f3d6732fc5b1b8b369108bdf6b --- /dev/null +++ b/physiological_integration/cases/case28/case28_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `24`, compare the 3-second baseline `[1449, 1452)` with the stimulus period `[1452, 1512)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_15.json b/physiological_integration/cases/case28/case28_15.json new file mode 100644 index 0000000000000000000000000000000000000000..96d59497ce48833ca6b15785f5653c0bc4e0d16b --- /dev/null +++ b/physiological_integration/cases/case28/case28_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `26`, compare the 3-second baseline `[1575, 1578)` with the stimulus period `[1578, 1638)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_16.json b/physiological_integration/cases/case28/case28_16.json new file mode 100644 index 0000000000000000000000000000000000000000..7125cf2c3e272b14edbbf058b8971d1cca932691 --- /dev/null +++ b/physiological_integration/cases/case28/case28_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `2`, compare the 3-second baseline `[63, 66)` with the stimulus period `[66, 126)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_17.json b/physiological_integration/cases/case28/case28_17.json new file mode 100644 index 0000000000000000000000000000000000000000..e76c5d329b0888f82498b731bbc7c02e6e62b8ce --- /dev/null +++ b/physiological_integration/cases/case28/case28_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `7`, compare the 3-second baseline `[378, 381)` with the stimulus period `[381, 441)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_18.json b/physiological_integration/cases/case28/case28_18.json new file mode 100644 index 0000000000000000000000000000000000000000..e3bf317f3c2991bbf8d6c1b29c2a8c876f467262 --- /dev/null +++ b/physiological_integration/cases/case28/case28_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `20`, compare the 3-second baseline `[1197, 1200)` with the stimulus period `[1200, 1260)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_19.json b/physiological_integration/cases/case28/case28_19.json new file mode 100644 index 0000000000000000000000000000000000000000..2276eceb6967b89431d65c4f723578c291574f5a --- /dev/null +++ b/physiological_integration/cases/case28/case28_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `30`, compare the 3-second baseline `[1827, 1830)` with the stimulus period `[1830, 1890)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_20.json b/physiological_integration/cases/case28/case28_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0a0cd9930261acb0e13b7bcf36fa907e3bf854f1 --- /dev/null +++ b/physiological_integration/cases/case28/case28_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `15`, compare the 3-second baseline `[882, 885)` with the stimulus period `[885, 945)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_21.json b/physiological_integration/cases/case28/case28_21.json new file mode 100644 index 0000000000000000000000000000000000000000..cbcca371ba2d38bddd259013c860709898e98c3f --- /dev/null +++ b/physiological_integration/cases/case28/case28_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `7`, compare the 3-second baseline `[378, 381)` with the stimulus period `[381, 441)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_22.json b/physiological_integration/cases/case28/case28_22.json new file mode 100644 index 0000000000000000000000000000000000000000..e1cb6f8599e275a3661cbfe4812d27f933eed10a --- /dev/null +++ b/physiological_integration/cases/case28/case28_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `11`, compare the 3-second baseline `[630, 633)` with the stimulus period `[633, 693)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_23.json b/physiological_integration/cases/case28/case28_23.json new file mode 100644 index 0000000000000000000000000000000000000000..9e850af83bc5904447b7256e27135ebcb00c2e4a --- /dev/null +++ b/physiological_integration/cases/case28/case28_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `25`, compare the 3-second baseline `[1512, 1515)` with the stimulus period `[1515, 1575)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_24.json b/physiological_integration/cases/case28/case28_24.json new file mode 100644 index 0000000000000000000000000000000000000000..3640e887433029abeb536118b0eb6ed102894fe4 --- /dev/null +++ b/physiological_integration/cases/case28/case28_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `2`, compare the 3-second baseline `[63, 66)` with the stimulus period `[66, 126)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_25.json b/physiological_integration/cases/case28/case28_25.json new file mode 100644 index 0000000000000000000000000000000000000000..4760d9aa59867aec1a84bdd56395dd111a388657 --- /dev/null +++ b/physiological_integration/cases/case28/case28_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `10`, compare the 3-second baseline `[567, 570)` with the stimulus period `[570, 630)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_26.json b/physiological_integration/cases/case28/case28_26.json new file mode 100644 index 0000000000000000000000000000000000000000..53e85ea4f26bb99a6022fe66f9efcae96f2eb739 --- /dev/null +++ b/physiological_integration/cases/case28/case28_26.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `33`, compare the 3-second baseline `[2016, 2019)` with the stimulus period `[2019, 2079)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_27.json b/physiological_integration/cases/case28/case28_27.json new file mode 100644 index 0000000000000000000000000000000000000000..d4e505b0e36fed9ca7359b4f7c5f9976fe394920 --- /dev/null +++ b/physiological_integration/cases/case28/case28_27.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `28`, compare the 3-second baseline `[1701, 1704)` with the stimulus period `[1704, 1764)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_28.json b/physiological_integration/cases/case28/case28_28.json new file mode 100644 index 0000000000000000000000000000000000000000..0f85298384921ff19c5af7597f1520c0d1651ce0 --- /dev/null +++ b/physiological_integration/cases/case28/case28_28.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `20`, compare the 3-second baseline `[1197, 1200)` with the stimulus period `[1200, 1260)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_29.json b/physiological_integration/cases/case28/case28_29.json new file mode 100644 index 0000000000000000000000000000000000000000..4022df4e2fe025ee111555739d005561ea9413d6 --- /dev/null +++ b/physiological_integration/cases/case28/case28_29.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `16`, compare the 3-second baseline `[945, 948)` with the stimulus period `[948, 1008)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case28/case28_30.json b/physiological_integration/cases/case28/case28_30.json new file mode 100644 index 0000000000000000000000000000000000000000..922c256f96b8a508e9d68f9a2d8f2242efa7c672 --- /dev/null +++ b/physiological_integration/cases/case28/case28_30.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case28_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `30`, compare the 3-second baseline `[1827, 1830)` with the stimulus period `[1830, 1890)`.\n\nCompare alpha power in `[8, 13)` Hz across posterior EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`. Report whether posterior alpha activity `increased` or `decreased` during stimulation.\n\nCompare the mean `GSR1` level between the same periods. Report whether the GSR level `increased` or `decreased` during stimulation.\n\nReport only `posterior_alpha_change` and `gsr_level_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- posterior_alpha_change\n- gsr_level_change\n\nEach value must be exactly `increased` or `decreased`.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"posterior_alpha_change\" and \"gsr_level_change\".\n4. Each value must be exactly \"increased\", \"decreased\", or null if missing or unclear.\n5. Interpret each direction as stimulus relative to baseline.\n\n### OUTPUT TEMPLATE\n{\"posterior_alpha_change\": \"\", \"gsr_level_change\": \"\"}", + "metrics": [ + { + "metric_id": "posterior_alpha_change", + "type": "categorical_check", + "target_key": "posterior_alpha_change", + "weight": 50, + "params": { + "gt_value": "increased" + } + }, + { + "metric_id": "gsr_level_change", + "type": "categorical_check", + "target_key": "gsr_level_change", + "weight": 50, + "params": { + "gt_value": "decreased" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_01.json b/physiological_integration/cases/case29/case29_01.json new file mode 100644 index 0000000000000000000000000000000000000000..fc0f7bcd5c9aa75cd022817bbf8f1926b80ce670 --- /dev/null +++ b/physiological_integration/cases/case29/case29_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [640, 820) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_02.json b/physiological_integration/cases/case29/case29_02.json new file mode 100644 index 0000000000000000000000000000000000000000..1bf0092b33c28f788f27785c91cbe23198cd0403 --- /dev/null +++ b/physiological_integration/cases/case29/case29_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [824, 1004) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_03.json b/physiological_integration/cases/case29/case29_03.json new file mode 100644 index 0000000000000000000000000000000000000000..08f77a4bc9d436b9eceb2b8cac223b1bc1a9575c --- /dev/null +++ b/physiological_integration/cases/case29/case29_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [1008, 1188) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_04.json b/physiological_integration/cases/case29/case29_04.json new file mode 100644 index 0000000000000000000000000000000000000000..7606ae1b97eafcbad552f58a30cab51d7d53460d --- /dev/null +++ b/physiological_integration/cases/case29/case29_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [1192, 1372) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_05.json b/physiological_integration/cases/case29/case29_05.json new file mode 100644 index 0000000000000000000000000000000000000000..08cf2f64e16660cadffb3d6ba9d1aea673800af5 --- /dev/null +++ b/physiological_integration/cases/case29/case29_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [1376, 1556) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_06.json b/physiological_integration/cases/case29/case29_06.json new file mode 100644 index 0000000000000000000000000000000000000000..17f183fa7eebf701707d1fc74882b93bc5ad116f --- /dev/null +++ b/physiological_integration/cases/case29/case29_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [1560, 1740) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_07.json b/physiological_integration/cases/case29/case29_07.json new file mode 100644 index 0000000000000000000000000000000000000000..d6310bf6bbf39e2bdeea5bf488d8dcfe226d9c68 --- /dev/null +++ b/physiological_integration/cases/case29/case29_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [6184, 6364) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_08.json b/physiological_integration/cases/case29/case29_08.json new file mode 100644 index 0000000000000000000000000000000000000000..7662dd4ff30438c76699fae7686dd4c8c9edb3ec --- /dev/null +++ b/physiological_integration/cases/case29/case29_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [1508, 1688) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_09.json b/physiological_integration/cases/case29/case29_09.json new file mode 100644 index 0000000000000000000000000000000000000000..34f6653a4d19a2627e5573373dab98cde04d7878 --- /dev/null +++ b/physiological_integration/cases/case29/case29_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [2112, 2292) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_10.json b/physiological_integration/cases/case29/case29_10.json new file mode 100644 index 0000000000000000000000000000000000000000..b3ce1e9b384129e02bc238e0f85e8aa01b61d782 --- /dev/null +++ b/physiological_integration/cases/case29/case29_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [2296, 2476) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_11.json b/physiological_integration/cases/case29/case29_11.json new file mode 100644 index 0000000000000000000000000000000000000000..fb3c7c4263a7605faebcf88ece554303f6c047b5 --- /dev/null +++ b/physiological_integration/cases/case29/case29_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [2480, 2660) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_12.json b/physiological_integration/cases/case29/case29_12.json new file mode 100644 index 0000000000000000000000000000000000000000..0fc2266e27fb0e024ebd3c6c708d2bbda820b512 --- /dev/null +++ b/physiological_integration/cases/case29/case29_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [2664, 2844) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_13.json b/physiological_integration/cases/case29/case29_13.json new file mode 100644 index 0000000000000000000000000000000000000000..b698ce68a7f3007de4b795df0d6703a54836e7f8 --- /dev/null +++ b/physiological_integration/cases/case29/case29_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [2848, 3028) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_14.json b/physiological_integration/cases/case29/case29_14.json new file mode 100644 index 0000000000000000000000000000000000000000..2c45b6c6e80cc3597c90c5ef87835b0c3880cb06 --- /dev/null +++ b/physiological_integration/cases/case29/case29_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [3032, 3212) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_15.json b/physiological_integration/cases/case29/case29_15.json new file mode 100644 index 0000000000000000000000000000000000000000..962d4d5110bf6e8dbcc9e442b051173f47da0893 --- /dev/null +++ b/physiological_integration/cases/case29/case29_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [3216, 3396) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_16.json b/physiological_integration/cases/case29/case29_16.json new file mode 100644 index 0000000000000000000000000000000000000000..16e0d080f2229a7afc0255930341f1fddc68b0b6 --- /dev/null +++ b/physiological_integration/cases/case29/case29_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [3400, 3580) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_17.json b/physiological_integration/cases/case29/case29_17.json new file mode 100644 index 0000000000000000000000000000000000000000..26e01c3d622520afc0b7652aaae3cea61470c552 --- /dev/null +++ b/physiological_integration/cases/case29/case29_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [3584, 3764) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_18.json b/physiological_integration/cases/case29/case29_18.json new file mode 100644 index 0000000000000000000000000000000000000000..c8aa04faf69289071529f7c26d280afe151a16da --- /dev/null +++ b/physiological_integration/cases/case29/case29_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [3768, 3948) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_19.json b/physiological_integration/cases/case29/case29_19.json new file mode 100644 index 0000000000000000000000000000000000000000..757e9edb137677da668586ac1d82fa4ac0fb8de1 --- /dev/null +++ b/physiological_integration/cases/case29/case29_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [3952, 4132) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_20.json b/physiological_integration/cases/case29/case29_20.json new file mode 100644 index 0000000000000000000000000000000000000000..2983bd46f8dd1c110f993258d42fb8d8e1c411cb --- /dev/null +++ b/physiological_integration/cases/case29/case29_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [4136, 4316) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_21.json b/physiological_integration/cases/case29/case29_21.json new file mode 100644 index 0000000000000000000000000000000000000000..0a3fabab342ab8084cbc7f45a4db7dc9fb4c6ea7 --- /dev/null +++ b/physiological_integration/cases/case29/case29_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [4320, 4500) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_22.json b/physiological_integration/cases/case29/case29_22.json new file mode 100644 index 0000000000000000000000000000000000000000..b874b7c1386a95bd0a41b4129e3aacc761eca89f --- /dev/null +++ b/physiological_integration/cases/case29/case29_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [4504, 4684) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_23.json b/physiological_integration/cases/case29/case29_23.json new file mode 100644 index 0000000000000000000000000000000000000000..274cc2157cb98612270409ffb3f1db7a26c416d4 --- /dev/null +++ b/physiological_integration/cases/case29/case29_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [1508, 1688) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_24.json b/physiological_integration/cases/case29/case29_24.json new file mode 100644 index 0000000000000000000000000000000000000000..13496630fde30144cac910eada6c1e1fc9000108 --- /dev/null +++ b/physiological_integration/cases/case29/case29_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [4872, 5052) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_25.json b/physiological_integration/cases/case29/case29_25.json new file mode 100644 index 0000000000000000000000000000000000000000..fcbd45036fd289121ae3cff1df84c9d3bea471a8 --- /dev/null +++ b/physiological_integration/cases/case29/case29_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [5056, 5236) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_26.json b/physiological_integration/cases/case29/case29_26.json new file mode 100644 index 0000000000000000000000000000000000000000..85fe22ae507b893093337f0c666b19da5c09ac7b --- /dev/null +++ b/physiological_integration/cases/case29/case29_26.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze [5240, 5420) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_27.json b/physiological_integration/cases/case29/case29_27.json new file mode 100644 index 0000000000000000000000000000000000000000..a58cbd7e30e95bbd32ed6060a610c8f8b474243b --- /dev/null +++ b/physiological_integration/cases/case29/case29_27.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze [5424, 5604) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_28.json b/physiological_integration/cases/case29/case29_28.json new file mode 100644 index 0000000000000000000000000000000000000000..2da140227e22a9483ee4cacfa8dabb50eee26b55 --- /dev/null +++ b/physiological_integration/cases/case29/case29_28.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze [1498, 1678) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "early" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_29.json b/physiological_integration/cases/case29/case29_29.json new file mode 100644 index 0000000000000000000000000000000000000000..da8d6798027f51a64541401336f799d39d29b708 --- /dev/null +++ b/physiological_integration/cases/case29/case29_29.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze [992, 1172) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + } + ] + } +} diff --git a/physiological_integration/cases/case29/case29_30.json b/physiological_integration/cases/case29/case29_30.json new file mode 100644 index 0000000000000000000000000000000000000000..60759e817adbb7cb1f2dbe62c9c74688ddcf570a --- /dev/null +++ b/physiological_integration/cases/case29/case29_30.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case29_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze [1176, 1356) seconds using six consecutive 30-second windows.\nFor each window, summarize EEG drowsiness evidence as the median across EEG channels of log(theta / (alpha + beta)), using theta [4, 8), alpha [8, 14), and beta [14, 31) Hz. Summarize EOG slow-eye-movement evidence as the mean [0.1, 1.0) Hz power across EOG-H and EOG-V.\nAverage windows 1–2 as early, 3–4 as middle, and 5–6 as late. For each modality, report the phase with the strongest evidence.\nReport only eeg_peak_phase and eog_peak_phase." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal drowsiness reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract:\n- eeg_peak_phase\n- eog_peak_phase\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown, explanations, or extra fields.\n3. The JSON keys must be EXACTLY \"eeg_peak_phase\" and \"eog_peak_phase\".\n4. Each value must be exactly \"early\", \"middle\", or \"late\", or null if missing or unclear.\n5. Extract the phase with the strongest evidence for each modality, not intermediate window results.\n\n### OUTPUT TEMPLATE\n{\"eeg_peak_phase\": \"\", \"eog_peak_phase\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_peak_phase", + "type": "categorical_check", + "target_key": "eeg_peak_phase", + "weight": 50, + "params": { + "gt_value": "middle" + } + }, + { + "metric_id": "eog_peak_phase", + "type": "categorical_check", + "target_key": "eog_peak_phase", + "weight": 50, + "params": { + "gt_value": "late" + } + } + ] + } +} diff --git a/physiological_integration/cases/case3/case3_01.json b/physiological_integration/cases/case3/case3_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d0244958f5d479c7f6237074166bf989a044e18e --- /dev/null +++ b/physiological_integration/cases/case3/case3_01.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "case03_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Find the start and end sample indices for channels `F3` and `GSR1` over the half-open interval `[0, 6)` seconds of the recording.\nReport only `channel_1_start_index`, `channel_1_end_index`, `channel_2_start_index`, and `channel_2_end_index`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- channel_1_start_index: Channel 1 start index.\n- channel_1_end_index: Channel 1 end index.\n- channel_2_start_index: Channel 2 start index.\n- channel_2_end_index: Channel 2 end index.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_start_index\": , \"channel_1_end_index\": , \"channel_2_start_index\": , \"channel_2_end_index\": }", + "metrics": [ + { + "metric_id": "channel_1_start_index", + "type": "numeric_check", + "target_key": "channel_1_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_1_end_index", + "type": "numeric_check", + "target_key": "channel_1_end_index", + "weight": 25, + "params": { + "gt_value": 3072.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_start_index", + "type": "numeric_check", + "target_key": "channel_2_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_end_index", + "type": "numeric_check", + "target_key": "channel_2_end_index", + "weight": 25, + "params": { + "gt_value": 3072.0, + "tolerance": 1.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case3/case3_06.json b/physiological_integration/cases/case3/case3_06.json new file mode 100644 index 0000000000000000000000000000000000000000..ef792a33d582807ca34939173efa841d433c1fb7 --- /dev/null +++ b/physiological_integration/cases/case3/case3_06.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "case03_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Find the start and end sample indices for channels `FT7` and `EOG-H` over the first 5 seconds (`[0, 5)`) of the recording.\nReport only `channel_1_start_index`, `channel_1_end_index`, `channel_2_start_index`, and `channel_2_end_index`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- channel_1_start_index: Channel 1 start index.\n- channel_1_end_index: Channel 1 end index.\n- channel_2_start_index: Channel 2 start index.\n- channel_2_end_index: Channel 2 end index.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_start_index\": , \"channel_1_end_index\": , \"channel_2_start_index\": , \"channel_2_end_index\": }", + "metrics": [ + { + "metric_id": "channel_1_start_index", + "type": "numeric_check", + "target_key": "channel_1_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_1_end_index", + "type": "numeric_check", + "target_key": "channel_1_end_index", + "weight": 25, + "params": { + "gt_value": 1000.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_start_index", + "type": "numeric_check", + "target_key": "channel_2_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_end_index", + "type": "numeric_check", + "target_key": "channel_2_end_index", + "weight": 25, + "params": { + "gt_value": 625.0, + "tolerance": 1.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case3/case3_11.json b/physiological_integration/cases/case3/case3_11.json new file mode 100644 index 0000000000000000000000000000000000000000..b893214e0ac4f4ea4c35e753e10fea25571d376f --- /dev/null +++ b/physiological_integration/cases/case3/case3_11.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "case03_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Find the start and end sample indices for channels `EEG(sec)` and `SaO2` over the half-open interval `[1110, 1140)` seconds of the recording.\nReport only `channel_1_start_index`, `channel_1_end_index`, `channel_2_start_index`, and `channel_2_end_index`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- channel_1_start_index: Channel 1 start index.\n- channel_1_end_index: Channel 1 end index.\n- channel_2_start_index: Channel 2 start index.\n- channel_2_end_index: Channel 2 end index.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_start_index\": , \"channel_1_end_index\": , \"channel_2_start_index\": , \"channel_2_end_index\": }", + "metrics": [ + { + "metric_id": "channel_1_start_index", + "type": "numeric_check", + "target_key": "channel_1_start_index", + "weight": 25, + "params": { + "gt_value": 138750.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_1_end_index", + "type": "numeric_check", + "target_key": "channel_1_end_index", + "weight": 25, + "params": { + "gt_value": 142500.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_start_index", + "type": "numeric_check", + "target_key": "channel_2_start_index", + "weight": 25, + "params": { + "gt_value": 1110.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_end_index", + "type": "numeric_check", + "target_key": "channel_2_end_index", + "weight": 25, + "params": { + "gt_value": 1140.0, + "tolerance": 1.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case3/case3_16.json b/physiological_integration/cases/case3/case3_16.json new file mode 100644 index 0000000000000000000000000000000000000000..3a774edab847c3e85d52aee3e2ade2e747e5dfd9 --- /dev/null +++ b/physiological_integration/cases/case3/case3_16.json @@ -0,0 +1,60 @@ +{ + "meta_info": { + "case_id": "case03_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Find the start and end sample indices for channels `FP1` and `S01D01_760` over the half-open interval `[0, 4.8)` seconds of the recording.\nReport only `channel_1_start_index`, `channel_1_end_index`, `channel_2_start_index`, and `channel_2_end_index`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- channel_1_start_index: Channel 1 start index.\n- channel_1_end_index: Channel 1 end index.\n- channel_2_start_index: Channel 2 start index.\n- channel_2_end_index: Channel 2 end index.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_start_index\": , \"channel_1_end_index\": , \"channel_2_start_index\": , \"channel_2_end_index\": }", + "metrics": [ + { + "metric_id": "channel_1_start_index", + "type": "numeric_check", + "target_key": "channel_1_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_1_end_index", + "type": "numeric_check", + "target_key": "channel_1_end_index", + "weight": 25, + "params": { + "gt_value": 4800.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_start_index", + "type": "numeric_check", + "target_key": "channel_2_start_index", + "weight": 25, + "params": { + "gt_value": 0.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_end_index", + "type": "numeric_check", + "target_key": "channel_2_end_index", + "weight": 25, + "params": { + "gt_value": 50.0, + "tolerance": 1.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case3/case3_21.json b/physiological_integration/cases/case3/case3_21.json new file mode 100644 index 0000000000000000000000000000000000000000..574ed2008759640443fdbaaa684c7672ea42e03c --- /dev/null +++ b/physiological_integration/cases/case3/case3_21.json @@ -0,0 +1,60 @@ +{ + "meta_info": { + "case_id": "case03_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Find the start and end sample indices for channels `FP1` and `GazePointX` over the half-open interval `[145, 150)` seconds of the recording.\nReport only `channel_1_start_index`, `channel_1_end_index`, `channel_2_start_index`, and `channel_2_end_index`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- channel_1_start_index: Channel 1 start index.\n- channel_1_end_index: Channel 1 end index.\n- channel_2_start_index: Channel 2 start index.\n- channel_2_end_index: Channel 2 end index.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"channel_1_start_index\": , \"channel_1_end_index\": , \"channel_2_start_index\": , \"channel_2_end_index\": }", + "metrics": [ + { + "metric_id": "channel_1_start_index", + "type": "numeric_check", + "target_key": "channel_1_start_index", + "weight": 25, + "params": { + "gt_value": 145000.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_1_end_index", + "type": "numeric_check", + "target_key": "channel_1_end_index", + "weight": 25, + "params": { + "gt_value": 150000.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_start_index", + "type": "numeric_check", + "target_key": "channel_2_start_index", + "weight": 25, + "params": { + "gt_value": 36250.0, + "tolerance": 1.0 + } + }, + { + "metric_id": "channel_2_end_index", + "type": "numeric_check", + "target_key": "channel_2_end_index", + "weight": 25, + "params": { + "gt_value": 37500.0, + "tolerance": 1.0 + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_01.json b/physiological_integration/cases/case30/case30_01.json new file mode 100644 index 0000000000000000000000000000000000000000..9d61ab90929e9d5dabf125622e4e2509a6df0fbb --- /dev/null +++ b/physiological_integration/cases/case30/case30_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `1`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_02.json b/physiological_integration/cases/case30/case30_02.json new file mode 100644 index 0000000000000000000000000000000000000000..67b55be16853bc975434bfbc324d0041bdb25865 --- /dev/null +++ b/physiological_integration/cases/case30/case30_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `2`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_03.json b/physiological_integration/cases/case30/case30_03.json new file mode 100644 index 0000000000000000000000000000000000000000..3375da976ca92e0b91a9456e21d372457610cdf1 --- /dev/null +++ b/physiological_integration/cases/case30/case30_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `13`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O2" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_04.json b/physiological_integration/cases/case30/case30_04.json new file mode 100644 index 0000000000000000000000000000000000000000..24327fc855c665ee6d37efb2f60334e0008daa15 --- /dev/null +++ b/physiological_integration/cases/case30/case30_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `4`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_05.json b/physiological_integration/cases/case30/case30_05.json new file mode 100644 index 0000000000000000000000000000000000000000..cd7882e3c04ed8fdb061e91412a0a83e4c82977f --- /dev/null +++ b/physiological_integration/cases/case30/case30_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `5`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_06.json b/physiological_integration/cases/case30/case30_06.json new file mode 100644 index 0000000000000000000000000000000000000000..2f21550fcdb7cbf19923b4ce4746472752ed1498 --- /dev/null +++ b/physiological_integration/cases/case30/case30_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `6`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_07.json b/physiological_integration/cases/case30/case30_07.json new file mode 100644 index 0000000000000000000000000000000000000000..292f321e124498d9809f9e027251fab57e47c5a2 --- /dev/null +++ b/physiological_integration/cases/case30/case30_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `7`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_08.json b/physiological_integration/cases/case30/case30_08.json new file mode 100644 index 0000000000000000000000000000000000000000..4d55c0a1624709c5293ab4793e7819f58399f54b --- /dev/null +++ b/physiological_integration/cases/case30/case30_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `18`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_09.json b/physiological_integration/cases/case30/case30_09.json new file mode 100644 index 0000000000000000000000000000000000000000..649bf8d9bb279ec200bf27f3e0e63c043e77b8a8 --- /dev/null +++ b/physiological_integration/cases/case30/case30_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `9`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_10.json b/physiological_integration/cases/case30/case30_10.json new file mode 100644 index 0000000000000000000000000000000000000000..e9fdd5d0a8bb8f028c9bcf84121811ca11916da6 --- /dev/null +++ b/physiological_integration/cases/case30/case30_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `10`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_5" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_11.json b/physiological_integration/cases/case30/case30_11.json new file mode 100644 index 0000000000000000000000000000000000000000..4a1e178138d47665ef97f2568c6668ba4df5256c --- /dev/null +++ b/physiological_integration/cases/case30/case30_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `11`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O2" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_12.json b/physiological_integration/cases/case30/case30_12.json new file mode 100644 index 0000000000000000000000000000000000000000..dd96d1895af08390aebf647787e2b5efeb7d7992 --- /dev/null +++ b/physiological_integration/cases/case30/case30_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `12`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_5" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_13.json b/physiological_integration/cases/case30/case30_13.json new file mode 100644 index 0000000000000000000000000000000000000000..07bd3450a780d71f8219241744739c47cfedd6f3 --- /dev/null +++ b/physiological_integration/cases/case30/case30_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `23`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O2" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_14.json b/physiological_integration/cases/case30/case30_14.json new file mode 100644 index 0000000000000000000000000000000000000000..229bca52ffa19ed252fb5b2ac863e7493b56f35c --- /dev/null +++ b/physiological_integration/cases/case30/case30_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `19`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O1" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_15.json b/physiological_integration/cases/case30/case30_15.json new file mode 100644 index 0000000000000000000000000000000000000000..1907fd8a5c088522b361933edcb9d47806e6463b --- /dev/null +++ b/physiological_integration/cases/case30/case30_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `15`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_4" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O2" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_16.json b/physiological_integration/cases/case30/case30_16.json new file mode 100644 index 0000000000000000000000000000000000000000..189fe0da0b76806ef9dfe0ea95b6ca0b5c76a0dc --- /dev/null +++ b/physiological_integration/cases/case30/case30_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `16`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_17.json b/physiological_integration/cases/case30/case30_17.json new file mode 100644 index 0000000000000000000000000000000000000000..3cb4585e1d42a7ef72c3874a5b1a51d366529a04 --- /dev/null +++ b/physiological_integration/cases/case30/case30_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `22`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_18.json b/physiological_integration/cases/case30/case30_18.json new file mode 100644 index 0000000000000000000000000000000000000000..414a26b6758d30e7b0865145133adaa7df5c4ca3 --- /dev/null +++ b/physiological_integration/cases/case30/case30_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `28`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O1" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_19.json b/physiological_integration/cases/case30/case30_19.json new file mode 100644 index 0000000000000000000000000000000000000000..009d129becf9bad4fba41cef6207f8324c4547ee --- /dev/null +++ b/physiological_integration/cases/case30/case30_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `24`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_3" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O1" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_20.json b/physiological_integration/cases/case30/case30_20.json new file mode 100644 index 0000000000000000000000000000000000000000..3c4992743b0eb19af7e881de59240ec15fce148a --- /dev/null +++ b/physiological_integration/cases/case30/case30_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `30`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_2" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_21.json b/physiological_integration/cases/case30/case30_21.json new file mode 100644 index 0000000000000000000000000000000000000000..015bed3f18b08363b24e29d65285a2b201de1309 --- /dev/null +++ b/physiological_integration/cases/case30/case30_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `21`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_22.json b/physiological_integration/cases/case30/case30_22.json new file mode 100644 index 0000000000000000000000000000000000000000..1a6f9207e2a361f9b89ac1d9e1947be4b0cc6758 --- /dev/null +++ b/physiological_integration/cases/case30/case30_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `27`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_23.json b/physiological_integration/cases/case30/case30_23.json new file mode 100644 index 0000000000000000000000000000000000000000..464c9b08ae6ee2afdde1fc35aadb07355d10abb7 --- /dev/null +++ b/physiological_integration/cases/case30/case30_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `36`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O1" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_24.json b/physiological_integration/cases/case30/case30_24.json new file mode 100644 index 0000000000000000000000000000000000000000..887477b985f75c42b52fcd8ae40ff0553e71bf28 --- /dev/null +++ b/physiological_integration/cases/case30/case30_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `29`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_3" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_25.json b/physiological_integration/cases/case30/case30_25.json new file mode 100644 index 0000000000000000000000000000000000000000..bf6bb2f4edcbdbd739ea2bcf6cdfc306a35ac37c --- /dev/null +++ b/physiological_integration/cases/case30/case30_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `35`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "O2" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_26.json b/physiological_integration/cases/case30/case30_26.json new file mode 100644 index 0000000000000000000000000000000000000000..e6d3a33d2d860808cf8a7d6bd46341a922a9830c --- /dev/null +++ b/physiological_integration/cases/case30/case30_26.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `26`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_27.json b/physiological_integration/cases/case30/case30_27.json new file mode 100644 index 0000000000000000000000000000000000000000..2e551eab6945eb7be3b49c7947b6b3b436e39faa --- /dev/null +++ b/physiological_integration/cases/case30/case30_27.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `32`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_28.json b/physiological_integration/cases/case30/case30_28.json new file mode 100644 index 0000000000000000000000000000000000000000..4c91ef13522e5c41abca545cd02561a7c0746946 --- /dev/null +++ b/physiological_integration/cases/case30/case30_28.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `19`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_29.json b/physiological_integration/cases/case30/case30_29.json new file mode 100644 index 0000000000000000000000000000000000000000..48c69a42ef21ae5dd0c7e23d85f0433e49be268c --- /dev/null +++ b/physiological_integration/cases/case30/case30_29.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `34`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_30.json b/physiological_integration/cases/case30/case30_30.json new file mode 100644 index 0000000000000000000000000000000000000000..4b4953999382d0d57c093adf873ccce031d2a9c3 --- /dev/null +++ b/physiological_integration/cases/case30/case30_30.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `14`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_31.json b/physiological_integration/cases/case30/case30_31.json new file mode 100644 index 0000000000000000000000000000000000000000..9c7d6b426efee40a7909e90f50f72c17f2cde25d --- /dev/null +++ b/physiological_integration/cases/case30/case30_31.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `31`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_32.json b/physiological_integration/cases/case30/case30_32.json new file mode 100644 index 0000000000000000000000000000000000000000..82067bcdb6bb1448247d38f54a5f42fe1a427bd4 --- /dev/null +++ b/physiological_integration/cases/case30/case30_32.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `37`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_33.json b/physiological_integration/cases/case30/case30_33.json new file mode 100644 index 0000000000000000000000000000000000000000..086ebac3fa56a8bf620a91873a29999a4de8fa20 --- /dev/null +++ b/physiological_integration/cases/case30/case30_33.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `37`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_34.json b/physiological_integration/cases/case30/case30_34.json new file mode 100644 index 0000000000000000000000000000000000000000..28671e422672ee865aa7c4cf789c25712e7598d0 --- /dev/null +++ b/physiological_integration/cases/case30/case30_34.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `39`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO3" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_35.json b/physiological_integration/cases/case30/case30_35.json new file mode 100644 index 0000000000000000000000000000000000000000..2e6c22f00e8f50ab5af24109313d39b2bbfd344b --- /dev/null +++ b/physiological_integration/cases/case30/case30_35.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `37`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_5" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_36.json b/physiological_integration/cases/case30/case30_36.json new file mode 100644 index 0000000000000000000000000000000000000000..6b5b85f9097685ab4b2b9853134135783a4ff2ce --- /dev/null +++ b/physiological_integration/cases/case30/case30_36.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "For DEAP trial `36`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_37.json b/physiological_integration/cases/case30/case30_37.json new file mode 100644 index 0000000000000000000000000000000000000000..c1f0d3a472bee083d79fed9ed1a5698f1f7481a9 --- /dev/null +++ b/physiological_integration/cases/case30/case30_37.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "For DEAP trial `9`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_4" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "P4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_38.json b/physiological_integration/cases/case30/case30_38.json new file mode 100644 index 0000000000000000000000000000000000000000..300fc2a91988cdf0d0f77c1c874312f4d6ab1f06 --- /dev/null +++ b/physiological_integration/cases/case30/case30_38.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "For DEAP trial `14`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_1" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_39.json b/physiological_integration/cases/case30/case30_39.json new file mode 100644 index 0000000000000000000000000000000000000000..35a9cad7cbc3e4368576e8e1ca74d95a3b39fbbc --- /dev/null +++ b/physiological_integration/cases/case30/case30_39.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "For DEAP trial `18`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_4" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case30/case30_40.json b/physiological_integration/cases/case30/case30_40.json new file mode 100644 index 0000000000000000000000000000000000000000..41831e217ca838ec69730c3d7d8e42f4fb0ca5fd --- /dev/null +++ b/physiological_integration/cases/case30/case30_40.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case30_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "For DEAP trial `6`, divide the 60-second stimulus period into six consecutive 10-second windows.\n\nFor `GSR1`, subtract the baseline mean from each window mean and report the window with the largest response as `peak_gsr_window`.\n\nWithin that window, compute alpha suppression in [8, 13) Hz for `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2` as baseline alpha power minus window alpha power. Report the channel with the largest suppression as `alpha_suppression_channel`.\n" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological reports.\n\n### TASK\nExtract the following fields:\n- peak_gsr_window: the 10-second stimulus window with the largest baseline-corrected GSR1 response.\n- alpha_suppression_channel: the posterior EEG channel with the largest alpha suppression within the selected GSR window.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `peak_gsr_window` and `alpha_suppression_channel`.\n4. `peak_gsr_window` must be one of `window_1`, `window_2`, `window_3`, `window_4`, `window_5`, `window_6`, or null.\n5. `alpha_suppression_channel` must be one of `P3`, `P4`, `PO3`, `PO4`, `O1`, `O2`, or null.\n6. Use null only when the result is missing or unclear.\n\n### OUTPUT TEMPLATE\n{\"peak_gsr_window\": \"\", \"alpha_suppression_channel\": \"\"}", + "metrics": [ + { + "metric_id": "peak_gsr_window", + "type": "categorical_check", + "target_key": "peak_gsr_window", + "weight": 50, + "params": { + "gt_value": "window_6" + } + }, + { + "metric_id": "alpha_suppression_channel", + "type": "categorical_check", + "target_key": "alpha_suppression_channel", + "weight": 50, + "params": { + "gt_value": "PO4" + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_01.json b/physiological_integration/cases/case31/case31_01.json new file mode 100644 index 0000000000000000000000000000000000000000..c86cac8ca6be890fc9dfc007119e90681907d8a1 --- /dev/null +++ b/physiological_integration/cases/case31/case31_01.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `2`, analyze the following two equal-duration, trial-relative 2-second windows: `[66, 68)` and `[187, 189)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_02.json b/physiological_integration/cases/case31/case31_02.json new file mode 100644 index 0000000000000000000000000000000000000000..5df809d17b157fea2828149ab00f3249fae83ff0 --- /dev/null +++ b/physiological_integration/cases/case31/case31_02.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `4`, analyze the following two equal-duration, trial-relative 2-second windows: `[13, 15)` and `[63, 65)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_03.json b/physiological_integration/cases/case31/case31_03.json new file mode 100644 index 0000000000000000000000000000000000000000..26550efc964de382a7effe336475d5c56215cf94 --- /dev/null +++ b/physiological_integration/cases/case31/case31_03.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `6`, analyze the following two equal-duration, trial-relative 2-second windows: `[2, 4)` and `[48, 50)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_04.json b/physiological_integration/cases/case31/case31_04.json new file mode 100644 index 0000000000000000000000000000000000000000..b7af8f29b8ae357fbcde34962d75bee369acf47b --- /dev/null +++ b/physiological_integration/cases/case31/case31_04.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `9`, analyze the following two equal-duration, trial-relative 2-second windows: `[129, 131)` and `[208, 210)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_05.json b/physiological_integration/cases/case31/case31_05.json new file mode 100644 index 0000000000000000000000000000000000000000..85d2ed5eeb4d0fe81602a40a595bae538c2f7d85 --- /dev/null +++ b/physiological_integration/cases/case31/case31_05.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `14`, analyze the following two equal-duration, trial-relative 2-second windows: `[141, 143)` and `[147, 149)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_06.json b/physiological_integration/cases/case31/case31_06.json new file mode 100644 index 0000000000000000000000000000000000000000..3deb9ea4a3e2d541a1f4cd86358a80f80e6b2388 --- /dev/null +++ b/physiological_integration/cases/case31/case31_06.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_01_session-20221001_1_events.npy", + "instruction": "For SEED-VII trial `19`, analyze the following two equal-duration, trial-relative 2-second windows: `[54, 56)` and `[6, 8)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_07.json b/physiological_integration/cases/case31/case31_07.json new file mode 100644 index 0000000000000000000000000000000000000000..5a3587b1ef9eecb8e81e28425c7b53424d205c0d --- /dev/null +++ b/physiological_integration/cases/case31/case31_07.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `6`, analyze the following two equal-duration, trial-relative 2-second windows: `[61, 63)` and `[135, 137)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_08.json b/physiological_integration/cases/case31/case31_08.json new file mode 100644 index 0000000000000000000000000000000000000000..27ffcb9cf0ac9f8140f2c30c613d20b0799ae870 --- /dev/null +++ b/physiological_integration/cases/case31/case31_08.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `13`, analyze the following two equal-duration, trial-relative 2-second windows: `[134, 136)` and `[18, 20)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_09.json b/physiological_integration/cases/case31/case31_09.json new file mode 100644 index 0000000000000000000000000000000000000000..961b40fa182b1da75d8728dc8739fa42fb79b57d --- /dev/null +++ b/physiological_integration/cases/case31/case31_09.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `19`, analyze the following two equal-duration, trial-relative 2-second windows: `[157, 159)` and `[4, 6)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_10.json b/physiological_integration/cases/case31/case31_10.json new file mode 100644 index 0000000000000000000000000000000000000000..ae1b08ace241f031cbc0013f85d14b16023f043d --- /dev/null +++ b/physiological_integration/cases/case31/case31_10.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `2`, analyze the following two equal-duration, trial-relative 2-second windows: `[173, 175)` and `[134, 136)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_11.json b/physiological_integration/cases/case31/case31_11.json new file mode 100644 index 0000000000000000000000000000000000000000..cbc7dea602faa72575f89f84666d2818095b9bd5 --- /dev/null +++ b/physiological_integration/cases/case31/case31_11.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `9`, analyze the following two equal-duration, trial-relative 2-second windows: `[159, 161)` and `[222, 224)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_12.json b/physiological_integration/cases/case31/case31_12.json new file mode 100644 index 0000000000000000000000000000000000000000..ccd8cc8553183206d44c5c0a6eb2a88064abcc83 --- /dev/null +++ b/physiological_integration/cases/case31/case31_12.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_02_session-20221011_1_events.npy", + "instruction": "For SEED-VII trial `17`, analyze the following two equal-duration, trial-relative 2-second windows: `[127, 129)` and `[115, 117)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "theta", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_13.json b/physiological_integration/cases/case31/case31_13.json new file mode 100644 index 0000000000000000000000000000000000000000..687edc99f7cc501e9f7e14d12a7b4cdb824c2b2c --- /dev/null +++ b/physiological_integration/cases/case31/case31_13.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `9`, analyze the following two equal-duration, trial-relative 2-second windows: `[35, 37)` and `[143, 145)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "beta", + "gamma", + "delta", + "theta", + "alpha" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "gamma", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_14.json b/physiological_integration/cases/case31/case31_14.json new file mode 100644 index 0000000000000000000000000000000000000000..2d53c61d13eb0788c71db2dc38683380b5099f04 --- /dev/null +++ b/physiological_integration/cases/case31/case31_14.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `10`, analyze the following two equal-duration, trial-relative 2-second windows: `[111, 113)` and `[142, 144)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_15.json b/physiological_integration/cases/case31/case31_15.json new file mode 100644 index 0000000000000000000000000000000000000000..6b3a9f793938a927e8c6154c2ba8550a4983b6be --- /dev/null +++ b/physiological_integration/cases/case31/case31_15.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `6`, analyze the following two equal-duration, trial-relative 2-second windows: `[2, 4)` and `[39, 41)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "gamma", + "theta", + "alpha" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "gamma", + "alpha", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_16.json b/physiological_integration/cases/case31/case31_16.json new file mode 100644 index 0000000000000000000000000000000000000000..da14f6f9eb82f5436392b186578fac11ac69cb80 --- /dev/null +++ b/physiological_integration/cases/case31/case31_16.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `7`, analyze the following two equal-duration, trial-relative 2-second windows: `[72, 74)` and `[105, 107)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "gamma", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_17.json b/physiological_integration/cases/case31/case31_17.json new file mode 100644 index 0000000000000000000000000000000000000000..c860e69b5de651df19839f35761b7129ca0debac --- /dev/null +++ b/physiological_integration/cases/case31/case31_17.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `2`, analyze the following two equal-duration, trial-relative 2-second windows: `[157, 159)` and `[168, 170)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "gamma", + "alpha", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "gamma", + "beta", + "delta", + "alpha", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_18.json b/physiological_integration/cases/case31/case31_18.json new file mode 100644 index 0000000000000000000000000000000000000000..0afd62b7a2afa560c6e87dc6fbe1c00e620e6449 --- /dev/null +++ b/physiological_integration/cases/case31/case31_18.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_03_session-20221009_1_events.npy", + "instruction": "For SEED-VII trial `4`, analyze the following two equal-duration, trial-relative 2-second windows: `[212, 214)` and `[62, 64)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "gamma", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "gamma", + "theta" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_19.json b/physiological_integration/cases/case31/case31_19.json new file mode 100644 index 0000000000000000000000000000000000000000..3620addeb33b00d5c8a5f0d45916bbbfd63125a9 --- /dev/null +++ b/physiological_integration/cases/case31/case31_19.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `2`, analyze the following two equal-duration, trial-relative 2-second windows: `[139, 141)` and `[201, 203)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_20.json b/physiological_integration/cases/case31/case31_20.json new file mode 100644 index 0000000000000000000000000000000000000000..ad999434132c84342e8dfc08104c017e0293cdd2 --- /dev/null +++ b/physiological_integration/cases/case31/case31_20.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `4`, analyze the following two equal-duration, trial-relative 2-second windows: `[50, 52)` and `[144, 146)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "alpha", + "delta", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_21.json b/physiological_integration/cases/case31/case31_21.json new file mode 100644 index 0000000000000000000000000000000000000000..365d253dbb61ba5ad6a287fef50cef61188f1cf7 --- /dev/null +++ b/physiological_integration/cases/case31/case31_21.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `6`, analyze the following two equal-duration, trial-relative 2-second windows: `[11, 13)` and `[99, 101)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_22.json b/physiological_integration/cases/case31/case31_22.json new file mode 100644 index 0000000000000000000000000000000000000000..5325ea889c94bbee00c27933a1d04b096c1622f7 --- /dev/null +++ b/physiological_integration/cases/case31/case31_22.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `12`, analyze the following two equal-duration, trial-relative 2-second windows: `[48, 50)` and `[37, 39)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "alpha", + "beta", + "delta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_23.json b/physiological_integration/cases/case31/case31_23.json new file mode 100644 index 0000000000000000000000000000000000000000..305d0898a1f223c67c65e240dcd5fd5067162672 --- /dev/null +++ b/physiological_integration/cases/case31/case31_23.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `16`, analyze the following two equal-duration, trial-relative 2-second windows: `[17, 19)` and `[115, 117)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "theta", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_24.json b/physiological_integration/cases/case31/case31_24.json new file mode 100644 index 0000000000000000000000000000000000000000..a29974b66b4457ce7a377f14296a49d7e2436535 --- /dev/null +++ b/physiological_integration/cases/case31/case31_24.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_04_session-20221030_1_events.npy", + "instruction": "For SEED-VII trial `19`, analyze the following two equal-duration, trial-relative 2-second windows: `[24, 26)` and `[220, 222)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "alpha", + "delta", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "alpha", + "beta", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_25.json b/physiological_integration/cases/case31/case31_25.json new file mode 100644 index 0000000000000000000000000000000000000000..3c627ed00bbd729c83e8e683d3c879b07e6ed026 --- /dev/null +++ b/physiological_integration/cases/case31/case31_25.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `2`, analyze the following two equal-duration, trial-relative 2-second windows: `[36, 38)` and `[201, 203)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_26.json b/physiological_integration/cases/case31/case31_26.json new file mode 100644 index 0000000000000000000000000000000000000000..4f19adc0c3e5599608d780e462e5ea3817faa7a6 --- /dev/null +++ b/physiological_integration/cases/case31/case31_26.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `6`, analyze the following two equal-duration, trial-relative 2-second windows: `[1, 3)` and `[75, 77)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_27.json b/physiological_integration/cases/case31/case31_27.json new file mode 100644 index 0000000000000000000000000000000000000000..30de6411c38f522cc688fc9f3f7ca41a06417609 --- /dev/null +++ b/physiological_integration/cases/case31/case31_27.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `10`, analyze the following two equal-duration, trial-relative 2-second windows: `[194, 196)` and `[31, 33)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "theta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_28.json b/physiological_integration/cases/case31/case31_28.json new file mode 100644 index 0000000000000000000000000000000000000000..fc38410c3cef21640636fe37fec0ff3ff5367b69 --- /dev/null +++ b/physiological_integration/cases/case31/case31_28.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `7`, analyze the following two equal-duration, trial-relative 2-second windows: `[133, 135)` and `[24, 26)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "beta", + "alpha", + "theta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_29.json b/physiological_integration/cases/case31/case31_29.json new file mode 100644 index 0000000000000000000000000000000000000000..cfde94bc5c3ddc877fef79eadc439bbba87c6663 --- /dev/null +++ b/physiological_integration/cases/case31/case31_29.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `9`, analyze the following two equal-duration, trial-relative 2-second windows: `[103, 105)` and `[131, 133)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case31/case31_30.json b/physiological_integration/cases/case31/case31_30.json new file mode 100644 index 0000000000000000000000000000000000000000..4002530a6f093711f5490930a4a583ecddd4b8b8 --- /dev/null +++ b/physiological_integration/cases/case31/case31_30.json @@ -0,0 +1,69 @@ +{ + "meta_info": { + "case_id": "case31_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "label_path": "data/multi/SEEDVII_05_session-20220929_1_events.npy", + "instruction": "For SEED-VII trial `19`, analyze the following two equal-duration, trial-relative 2-second windows: `[97, 99)` and `[65, 67)` seconds.\n\nClassify a window as `central` when at least half of its valid gaze samples fall within the central 50% × 50% screen region; otherwise classify it as `peripheral`. Ignore invalid gaze samples. These two windows were selected to contain one central and one peripheral window.\n\nFor each window, compute `delta`, `theta`, `alpha`, `beta`, and `gamma` power across `P3`, `PZ`, `P4`, `PO3`, `POZ`, `PO4`, `O1`, `OZ`, and `O2`. Convert the five band powers within each window into proportions of their total power.\n\nAverage the band proportions separately for the central and peripheral windows, then rank the five bands from highest to lowest proportion.\n\nReport `central_band_ranking` and `peripheral_band_ranking`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for EEG band-ranking results.\n\n### TASK\nExtract:\n- `central_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the central-gaze window.\n- `peripheral_band_ranking`: the five EEG bands ranked from highest to lowest mean proportion across the peripheral-gaze window.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. The keys must be exactly `central_band_ranking` and `peripheral_band_ranking`.\n3. Each value must be a JSON array containing the five bands in ranked order.\n4. The allowed band names are exactly `delta`, `theta`, `alpha`, `beta`, and `gamma`.\n5. Each band must appear exactly once in each array.\n6. Do not output Markdown, explanations, proportions, units, or extra fields.\n7. Return null for a ranking that is missing, incomplete, duplicated, or unclear.\n\n### OUTPUT TEMPLATE\n{\"central_band_ranking\": [\"\", \"\", \"\", \"\", \"\"], \"peripheral_band_ranking\": [\"\", \"\", \"\", \"\", \"\"]}", + "metrics": [ + { + "metric_id": "central_band_ranking", + "type": "sequence_match_check", + "target_key": "central_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "beta", + "alpha", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + }, + { + "metric_id": "peripheral_band_ranking", + "type": "sequence_match_check", + "target_key": "peripheral_band_ranking", + "weight": 50, + "params": { + "gt_value": [ + "delta", + "theta", + "alpha", + "beta", + "gamma" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ] + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_01.json b/physiological_integration/cases/case32/case32_01.json new file mode 100644 index 0000000000000000000000000000000000000000..8f520a60077f0061cf0ed1c5629292c7dc65e036 --- /dev/null +++ b/physiological_integration/cases/case32/case32_01.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[46.447, 115.403)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_02.json b/physiological_integration/cases/case32/case32_02.json new file mode 100644 index 0000000000000000000000000000000000000000..371becf20970f097ed0cc7cb45148030b113b2ce --- /dev/null +++ b/physiological_integration/cases/case32/case32_02.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[184.307, 253.202)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_03.json b/physiological_integration/cases/case32/case32_03.json new file mode 100644 index 0000000000000000000000000000000000000000..daca6f6d9ecfa0b3c93cedc9225dd27bb4d3b6c0 --- /dev/null +++ b/physiological_integration/cases/case32/case32_03.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[253.202, 322.06)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_04.json b/physiological_integration/cases/case32/case32_04.json new file mode 100644 index 0000000000000000000000000000000000000000..51c49b1517327e3c4d013976f106169b0aa6ebd2 --- /dev/null +++ b/physiological_integration/cases/case32/case32_04.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[322.06, 391.201)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_05.json b/physiological_integration/cases/case32/case32_05.json new file mode 100644 index 0000000000000000000000000000000000000000..3cee25c587579761bb519047f88a62cb0eb96930 --- /dev/null +++ b/physiological_integration/cases/case32/case32_05.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[391.201, 496)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_06.json b/physiological_integration/cases/case32/case32_06.json new file mode 100644 index 0000000000000000000000000000000000000000..e0b1faaa5dfda367c2fe7c9cb51c7dc6b4790646 --- /dev/null +++ b/physiological_integration/cases/case32/case32_06.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[46.447, 184.307)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_07.json b/physiological_integration/cases/case32/case32_07.json new file mode 100644 index 0000000000000000000000000000000000000000..f18308eb33d7a3a47e23cace7fc78657aed355e5 --- /dev/null +++ b/physiological_integration/cases/case32/case32_07.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[115.403, 253.202)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_08.json b/physiological_integration/cases/case32/case32_08.json new file mode 100644 index 0000000000000000000000000000000000000000..a6398b72694d93f534cd6ea31465fcb8260630d3 --- /dev/null +++ b/physiological_integration/cases/case32/case32_08.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[184.307, 322.06)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_09.json b/physiological_integration/cases/case32/case32_09.json new file mode 100644 index 0000000000000000000000000000000000000000..1882366a995dc26cb62bcc86a1741e8af3d6ceee --- /dev/null +++ b/physiological_integration/cases/case32/case32_09.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[44.039, 113.026)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_10.json b/physiological_integration/cases/case32/case32_10.json new file mode 100644 index 0000000000000000000000000000000000000000..b0c78bec9f2e76c8cf5e0b270d91c76608f0022a --- /dev/null +++ b/physiological_integration/cases/case32/case32_10.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[182.136, 250.836)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_11.json b/physiological_integration/cases/case32/case32_11.json new file mode 100644 index 0000000000000000000000000000000000000000..814a99f4c44c97909b1a7cf90aada26cd7456028 --- /dev/null +++ b/physiological_integration/cases/case32/case32_11.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[250.836, 319.803)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_12.json b/physiological_integration/cases/case32/case32_12.json new file mode 100644 index 0000000000000000000000000000000000000000..80424638472bfeed59329d53d37d55c897bfb6cb --- /dev/null +++ b/physiological_integration/cases/case32/case32_12.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[319.803, 388.756)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_13.json b/physiological_integration/cases/case32/case32_13.json new file mode 100644 index 0000000000000000000000000000000000000000..3f730f46cfdee0a556ed4bf49665a53928b3508a --- /dev/null +++ b/physiological_integration/cases/case32/case32_13.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[388.756, 492)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_14.json b/physiological_integration/cases/case32/case32_14.json new file mode 100644 index 0000000000000000000000000000000000000000..3456609dde8943b6bf32fd73a073f8e100cde7c6 --- /dev/null +++ b/physiological_integration/cases/case32/case32_14.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[44.039, 182.136)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_15.json b/physiological_integration/cases/case32/case32_15.json new file mode 100644 index 0000000000000000000000000000000000000000..ed1f7c3ef3cfdad13b5bc6b61246b82e0375af6e --- /dev/null +++ b/physiological_integration/cases/case32/case32_15.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[113.026, 250.836)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_16.json b/physiological_integration/cases/case32/case32_16.json new file mode 100644 index 0000000000000000000000000000000000000000..5e7ee536352cebd726b54f85414a362456dfe3a1 --- /dev/null +++ b/physiological_integration/cases/case32/case32_16.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[182.136, 319.803)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_17.json b/physiological_integration/cases/case32/case32_17.json new file mode 100644 index 0000000000000000000000000000000000000000..a10d7095644a317c8526611c3c6778f7a4c3777d --- /dev/null +++ b/physiological_integration/cases/case32/case32_17.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[42.64, 110.499)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_18.json b/physiological_integration/cases/case32/case32_18.json new file mode 100644 index 0000000000000000000000000000000000000000..a04898581094f96b59a51ff8323d835fae086486 --- /dev/null +++ b/physiological_integration/cases/case32/case32_18.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[178.371, 246.14)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_19.json b/physiological_integration/cases/case32/case32_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5c7967f7aa7509a61833fd70f324bb39848c2177 --- /dev/null +++ b/physiological_integration/cases/case32/case32_19.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[246.14, 314.092)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_20.json b/physiological_integration/cases/case32/case32_20.json new file mode 100644 index 0000000000000000000000000000000000000000..902e1dfee291771d5952de3aada0c2ce606ea230 --- /dev/null +++ b/physiological_integration/cases/case32/case32_20.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[314.092, 381.907)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_21.json b/physiological_integration/cases/case32/case32_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ef2ef29a4652770937f661f6597d3216b2e2a895 --- /dev/null +++ b/physiological_integration/cases/case32/case32_21.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[381.907, 485)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_22.json b/physiological_integration/cases/case32/case32_22.json new file mode 100644 index 0000000000000000000000000000000000000000..3ca9fa14281457fb2325486f389321f0c702bda8 --- /dev/null +++ b/physiological_integration/cases/case32/case32_22.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[42.64, 178.371)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_23.json b/physiological_integration/cases/case32/case32_23.json new file mode 100644 index 0000000000000000000000000000000000000000..8b3b46b2e13c5548e48a38283c01efc6ede5c833 --- /dev/null +++ b/physiological_integration/cases/case32/case32_23.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[246.14, 381.907)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_24.json b/physiological_integration/cases/case32/case32_24.json new file mode 100644 index 0000000000000000000000000000000000000000..5e5dbebe04751926693726f89cfffc2d32a69680 --- /dev/null +++ b/physiological_integration/cases/case32/case32_24.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[178.371, 314.092)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_25.json b/physiological_integration/cases/case32/case32_25.json new file mode 100644 index 0000000000000000000000000000000000000000..7c482e2132f8c59ae955f78b30759fb5ff9f79b2 --- /dev/null +++ b/physiological_integration/cases/case32/case32_25.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[42.083, 109.709)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_26.json b/physiological_integration/cases/case32/case32_26.json new file mode 100644 index 0000000000000000000000000000000000000000..f3d514275a843b7fd24941f72fd6b3dbf763768a --- /dev/null +++ b/physiological_integration/cases/case32/case32_26.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[177.37, 245.18)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_27.json b/physiological_integration/cases/case32/case32_27.json new file mode 100644 index 0000000000000000000000000000000000000000..ffb43a88c5eaf45b5c1694f41b38c6eea49a8670 --- /dev/null +++ b/physiological_integration/cases/case32/case32_27.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[245.18, 312.919)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_28.json b/physiological_integration/cases/case32/case32_28.json new file mode 100644 index 0000000000000000000000000000000000000000..bfb0c244a0c4f8fe09d163127d9787756ce29fad --- /dev/null +++ b/physiological_integration/cases/case32/case32_28.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[312.919, 380.747)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_29.json b/physiological_integration/cases/case32/case32_29.json new file mode 100644 index 0000000000000000000000000000000000000000..bdf082cd81102707185c93539eb101cedf0d857c --- /dev/null +++ b/physiological_integration/cases/case32/case32_29.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[380.747, 482)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_30.json b/physiological_integration/cases/case32/case32_30.json new file mode 100644 index 0000000000000000000000000000000000000000..d950b62016e69ca53f7f2ee9b0a589c8f0efe438 --- /dev/null +++ b/physiological_integration/cases/case32/case32_30.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[42.083, 177.37)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_31.json b/physiological_integration/cases/case32/case32_31.json new file mode 100644 index 0000000000000000000000000000000000000000..0b6fc87dbac2c408050c86bd1af5c06d58790f8b --- /dev/null +++ b/physiological_integration/cases/case32/case32_31.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[109.709, 245.18)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_32.json b/physiological_integration/cases/case32/case32_32.json new file mode 100644 index 0000000000000000000000000000000000000000..dfdef41ae9415fde73105139de771ef4e2625a58 --- /dev/null +++ b/physiological_integration/cases/case32/case32_32.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[177.37, 312.919)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_33.json b/physiological_integration/cases/case32/case32_33.json new file mode 100644 index 0000000000000000000000000000000000000000..53334cfe0d3f2e4bb195ee8118c43ff348c3c73b --- /dev/null +++ b/physiological_integration/cases/case32/case32_33.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[62.274, 129.693)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_34.json b/physiological_integration/cases/case32/case32_34.json new file mode 100644 index 0000000000000000000000000000000000000000..4631fc430b3173f0436d6951f150031117f429a5 --- /dev/null +++ b/physiological_integration/cases/case32/case32_34.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[197.086, 264.745)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_35.json b/physiological_integration/cases/case32/case32_35.json new file mode 100644 index 0000000000000000000000000000000000000000..bfc4e83b1b99b900ff918f682af947af32e8f9d6 --- /dev/null +++ b/physiological_integration/cases/case32/case32_35.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[264.745, 332.247)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_36.json b/physiological_integration/cases/case32/case32_36.json new file mode 100644 index 0000000000000000000000000000000000000000..dd8aea3370d6f3644e477438df1c84ec49935c6b --- /dev/null +++ b/physiological_integration/cases/case32/case32_36.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[332.247, 399.657)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_37.json b/physiological_integration/cases/case32/case32_37.json new file mode 100644 index 0000000000000000000000000000000000000000..7799a5c82dbe281be01622ee42c91f176ebfe4b2 --- /dev/null +++ b/physiological_integration/cases/case32/case32_37.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[399.657, 504)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_38.json b/physiological_integration/cases/case32/case32_38.json new file mode 100644 index 0000000000000000000000000000000000000000..6fb7b4b73e765a4e2ac9cd286922b0c5982394b5 --- /dev/null +++ b/physiological_integration/cases/case32/case32_38.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[62.274, 197.086)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for Go.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors Go.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_39.json b/physiological_integration/cases/case32/case32_39.json new file mode 100644 index 0000000000000000000000000000000000000000..b736ec5f7799fb78d8e12a7fb0ada3932d3d8e27 --- /dev/null +++ b/physiological_integration/cases/case32/case32_39.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[129.693, 264.745)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for NoGo.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor the same condition.\n\nA correct response must:\n1. State or clearly imply that EEG favors NoGo.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor the same condition.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case32/case32_40.json b/physiological_integration/cases/case32/case32_40.json new file mode 100644 index 0000000000000000000000000000000000000000..da0e85ed27d06a408014e8890f94580888bcaaba --- /dev/null +++ b/physiological_integration/cases/case32/case32_40.json @@ -0,0 +1,48 @@ +{ + "meta_info": { + "case_id": "case32_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Compare `Go` (event code `16`) and `NoGo` (event code `32`) events within `[197.086, 332.247)` seconds, using only events with complete EEG and fNIRS epochs.\n\nFor EEG channels `FP1`, `AFz`, and `F1`, compare the baseline-corrected theta power in `[4, 8)` Hz between the two conditions, using `[-1, 0)` seconds as baseline and `[0, 1)` seconds as response. Report the stronger condition as `eeg_stronger_condition`.\n\nFor fNIRS pair `S03D02`, compare the baseline-corrected HbO response between the two conditions, using `[-2, 0)` seconds as baseline and `[4, 8)` seconds as response. Report the stronger condition as `hbo_stronger_condition`.\n\nBriefly state whether EEG and HbO favor the same condition or different conditions. Treat this as descriptive cross-modal evidence without making causal or inhibition-specific claims." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal neurophysiological reports.\n\n### TASK\nExtract:\n- eeg_stronger_condition: the condition with the stronger EEG theta response.\n- hbo_stronger_condition: the condition with the stronger HbO response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, or extra fields.\n3. The keys must be exactly `eeg_stronger_condition` and `hbo_stronger_condition`.\n4. Each value must be exactly `Go`, `NoGo`, or null if missing or unclear.\n5. Do not infer a missing result from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"eeg_stronger_condition\": \"\", \"hbo_stronger_condition\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_stronger_condition", + "type": "categorical_check", + "target_key": "eeg_stronger_condition", + "weight": 30, + "params": { + "gt_value": "Go" + } + }, + { + "metric_id": "hbo_stronger_condition", + "type": "categorical_check", + "target_key": "hbo_stronger_condition", + "weight": 30, + "params": { + "gt_value": "NoGo" + } + }, + { + "metric_id": "cross_modal_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the response correctly interprets the joint EEG and HbO condition pattern.\n\nGround truth:\n- EEG theta response is stronger for Go.\n- HbO response is stronger for NoGo.\n- Therefore, the two modalities favor different conditions.\n\nA correct response must:\n1. State or clearly imply that EEG favors Go.\n2. State or clearly imply that HbO favors NoGo.\n3. State or clearly imply that the modalities favor different conditions.\n4. Treat the result as descriptive cross-modal evidence across repeated events.\n5. Avoid claiming proof of response inhibition, a causal EEG-fNIRS relationship, statistical significance, or a clinical mechanism.\n6. Not contradict either reported categorical result.\n\nExact wording is not required. Return {\"status\": true} only when the interpretation satisfies all requirements. Return {\"status\": false} otherwise.\nOutput exactly one JSON object with the key status and a Boolean value. Do not output explanations, Markdown, scores, or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_01.json b/physiological_integration/cases/case33/case33_01.json new file mode 100644 index 0000000000000000000000000000000000000000..695221d5618a39e3cf5fc7a17c94fd9a60aa12a9 --- /dev/null +++ b/physiological_integration/cases/case33/case33_01.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [46.447, 115.403) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.000000000000001, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 3.000000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_02.json b/physiological_integration/cases/case33/case33_02.json new file mode 100644 index 0000000000000000000000000000000000000000..4880467fd82089b3078f9e2d65ac3e75eaac94c3 --- /dev/null +++ b/physiological_integration/cases/case33/case33_02.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [44.039, 113.026) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.8000000000000016, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.950000 seconds;\n- hbo_peak_latency_sec = 3.800000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_03.json b/physiological_integration/cases/case33/case33_03.json new file mode 100644 index 0000000000000000000000000000000000000000..f0fee80caf666aec25b7a14330d91953da96e24b --- /dev/null +++ b/physiological_integration/cases/case33/case33_03.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [110.499, 178.371) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 4.400000000000002, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.800000 seconds;\n- hbo_peak_latency_sec = 4.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_04.json b/physiological_integration/cases/case33/case33_04.json new file mode 100644 index 0000000000000000000000000000000000000000..6dd645bc98477aa05f5abc30194fddf56aaec5e4 --- /dev/null +++ b/physiological_integration/cases/case33/case33_04.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [177.37, 245.18) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.400000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 8.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_05.json b/physiological_integration/cases/case33/case33_05.json new file mode 100644 index 0000000000000000000000000000000000000000..172bbc54eba10a4d018226fa2e3ec87800fd9dba --- /dev/null +++ b/physiological_integration/cases/case33/case33_05.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [62.274, 98.085) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.5, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.600000000000007, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.500000 seconds;\n- hbo_peak_latency_sec = 9.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_06.json b/physiological_integration/cases/case33/case33_06.json new file mode 100644 index 0000000000000000000000000000000000000000..532e8e0898d2e95cb091c4ccb5b8ab2eb1b503a4 --- /dev/null +++ b/physiological_integration/cases/case33/case33_06.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [322.06, 391.201) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 2.4000000000000004, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.950000 seconds;\n- hbo_peak_latency_sec = 2.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_07.json b/physiological_integration/cases/case33/case33_07.json new file mode 100644 index 0000000000000000000000000000000000000000..bd9f99554cd67d0668999914bf9f05f6122f7cb6 --- /dev/null +++ b/physiological_integration/cases/case33/case33_07.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [113.026, 182.136) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.5, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.400000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.500000 seconds;\n- hbo_peak_latency_sec = 8.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_08.json b/physiological_integration/cases/case33/case33_08.json new file mode 100644 index 0000000000000000000000000000000000000000..56b59f7e7c0a84241d978e6fb127a2db69ad6b18 --- /dev/null +++ b/physiological_integration/cases/case33/case33_08.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [381.907, 484.8) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.600000000000005, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 8.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_09.json b/physiological_integration/cases/case33/case33_09.json new file mode 100644 index 0000000000000000000000000000000000000000..5e0dcef0d94901dfe24d42a322181a8e0e70caad --- /dev/null +++ b/physiological_integration/cases/case33/case33_09.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [245.18, 312.919) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 4.400000000000002, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.950000 seconds;\n- hbo_peak_latency_sec = 4.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_10.json b/physiological_integration/cases/case33/case33_10.json new file mode 100644 index 0000000000000000000000000000000000000000..d000d99bfb05ec4154a964e0c7c3841a6c142ba4 --- /dev/null +++ b/physiological_integration/cases/case33/case33_10.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [66.76, 102.394) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.2, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.400000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.200000 seconds;\n- hbo_peak_latency_sec = 9.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_11.json b/physiological_integration/cases/case33/case33_11.json new file mode 100644 index 0000000000000000000000000000000000000000..f55f9ab3e35644f46c9f3f1120f861cd23c30613 --- /dev/null +++ b/physiological_integration/cases/case33/case33_11.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [391.201, 496) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.6000000000000014, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.950000 seconds;\n- hbo_peak_latency_sec = 3.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_12.json b/physiological_integration/cases/case33/case33_12.json new file mode 100644 index 0000000000000000000000000000000000000000..59acba1a4ea788ebeaf57c92b11cc629e8a5b791 --- /dev/null +++ b/physiological_integration/cases/case33/case33_12.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [182.136, 250.836) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.6000000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.800000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.600000 seconds;\n- hbo_peak_latency_sec = 8.800000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_13.json b/physiological_integration/cases/case33/case33_13.json new file mode 100644 index 0000000000000000000000000000000000000000..edfb7295e38aaf086d4372c193d9738594c9f3ce --- /dev/null +++ b/physiological_integration/cases/case33/case33_13.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [55.837, 87.375) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.600000000000005, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 8.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_14.json b/physiological_integration/cases/case33/case33_14.json new file mode 100644 index 0000000000000000000000000000000000000000..2f472290eb5f2cc768678ed92c942a27b8c45ffb --- /dev/null +++ b/physiological_integration/cases/case33/case33_14.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [312.919, 380.747) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.200000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.800000 seconds;\n- hbo_peak_latency_sec = 8.200000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_15.json b/physiological_integration/cases/case33/case33_15.json new file mode 100644 index 0000000000000000000000000000000000000000..5722dc7e2c80789f02ebec22df411793fdb294ff --- /dev/null +++ b/physiological_integration/cases/case33/case33_15.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [66.76, 108.8) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.2, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.600000000000007, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.200000 seconds;\n- hbo_peak_latency_sec = 9.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_16.json b/physiological_integration/cases/case33/case33_16.json new file mode 100644 index 0000000000000000000000000000000000000000..e119e0d1e0dcdb504c4e71a39cf275e9c053d792 --- /dev/null +++ b/physiological_integration/cases/case33/case33_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [46.757, 87.678) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.000000000000001, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 3.000000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_17.json b/physiological_integration/cases/case33/case33_17.json new file mode 100644 index 0000000000000000000000000000000000000000..a317fe96a019eecc1a68aed7fa5f1f1ee90a39af --- /dev/null +++ b/physiological_integration/cases/case33/case33_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [388.756, 492) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.45, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.800000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.450000 seconds;\n- hbo_peak_latency_sec = 8.800000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_18.json b/physiological_integration/cases/case33/case33_18.json new file mode 100644 index 0000000000000000000000000000000000000000..20974c9daf7824e6367778a1ec532b3ad7cb6a4a --- /dev/null +++ b/physiological_integration/cases/case33/case33_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [55.837, 91.661) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 8.600000000000005, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 8.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_19.json b/physiological_integration/cases/case33/case33_19.json new file mode 100644 index 0000000000000000000000000000000000000000..23db7caa3523de5ff90acf9f11b0f33c45160d2d --- /dev/null +++ b/physiological_integration/cases/case33/case33_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [50.917, 121.937) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 2.4000000000000004, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 2.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_20.json b/physiological_integration/cases/case33/case33_20.json new file mode 100644 index 0000000000000000000000000000000000000000..98a61bb0cd7ace434ddcbc2c30a843dd83047888 --- /dev/null +++ b/physiological_integration/cases/case33/case33_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [66.76, 113.075) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.2, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.600000000000007, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.200000 seconds;\n- hbo_peak_latency_sec = 9.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_21.json b/physiological_integration/cases/case33/case33_21.json new file mode 100644 index 0000000000000000000000000000000000000000..6b5a3ea860d44eb105cf59a03f45e175aa533fad --- /dev/null +++ b/physiological_integration/cases/case33/case33_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [57.816, 94.39) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 2.6000000000000005, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 2.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_22.json b/physiological_integration/cases/case33/case33_22.json new file mode 100644 index 0000000000000000000000000000000000000000..7dfe6eb4753f48ef7d5943dfb81cfcfd21187ba9 --- /dev/null +++ b/physiological_integration/cases/case33/case33_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [44.359, 78.609) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.6000000000000014, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 3.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_23.json b/physiological_integration/cases/case33/case33_23.json new file mode 100644 index 0000000000000000000000000000000000000000..1c73a7d489e30ac965faa148fe61b2a57c3e3d5d --- /dev/null +++ b/physiological_integration/cases/case33/case33_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [60.198, 124.922) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 4.200000000000002, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 4.200000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_24.json b/physiological_integration/cases/case33/case33_24.json new file mode 100644 index 0000000000000000000000000000000000000000..834459fb923dfb523f5d54d10e509e7a8bc63454 --- /dev/null +++ b/physiological_integration/cases/case33/case33_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [65.971, 121.937) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 2.6000000000000005, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 2.600000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_25.json b/physiological_integration/cases/case33/case33_25.json new file mode 100644 index 0000000000000000000000000000000000000000..045b92ee2732b67655ce9bc5d4301f4c7d588174 --- /dev/null +++ b/physiological_integration/cases/case33/case33_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [162.073, 224.382) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.2, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.200000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.200000 seconds;\n- hbo_peak_latency_sec = 9.200000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_26.json b/physiological_integration/cases/case33/case33_26.json new file mode 100644 index 0000000000000000000000000000000000000000..9f89e6864b51417db73c70a7e77ebe3c8d70d22f --- /dev/null +++ b/physiological_integration/cases/case33/case33_26.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [46.757, 94.39) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8500000000000001, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.000000000000001, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.850000 seconds;\n- hbo_peak_latency_sec = 3.000000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_27.json b/physiological_integration/cases/case33/case33_27.json new file mode 100644 index 0000000000000000000000000000000000000000..ee418eadf2123e6aaca55b655fddbf16525571bb --- /dev/null +++ b/physiological_integration/cases/case33/case33_27.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [44.359, 83.021) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 3.4000000000000012, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 3.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_28.json b/physiological_integration/cases/case33/case33_28.json new file mode 100644 index 0000000000000000000000000000000000000000..5787d5ccff8b6c6e8e6ca1e698037b62daa5a5fd --- /dev/null +++ b/physiological_integration/cases/case33/case33_28.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [66.68, 124.922) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.8, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 4.400000000000002, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.800000 seconds;\n- hbo_peak_latency_sec = 4.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_29.json b/physiological_integration/cases/case33/case33_29.json new file mode 100644 index 0000000000000000000000000000000000000000..e49f548ce8932afa48b44bc2c6b7c1a9146c39f5 --- /dev/null +++ b/physiological_integration/cases/case33/case33_29.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [50.917, 126.275) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.9, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 2.4000000000000004, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.900000 seconds;\n- hbo_peak_latency_sec = 2.400000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case33/case33_30.json b/physiological_integration/cases/case33/case33_30.json new file mode 100644 index 0000000000000000000000000000000000000000..00d3f6f5029640f22ca2d27102308697a6d19ca1 --- /dev/null +++ b/physiological_integration/cases/case33/case33_30.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case33_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Using complete `NoGo` events (event code `32`) within [132.011, 230.863) seconds, estimate the event-averaged frontal theta response from `FP1`, `AFz`, and `F1`, and the prefrontal HbO response from `S03D02`.\n\nBaseline-correct theta using `[-1, 0)` seconds and find its peak latency within `[0, 2)` seconds. Baseline-correct HbO using `[-2, 0)` seconds and find its peak latency within `[2, 10)` seconds.\n\nReport both latencies relative to event onset as `theta_peak_latency_sec` and `hbo_peak_latency_sec`. Briefly describe whether the delayed HbO peak follows the early theta peak. Treat this as an event-averaged temporal association, not proof of causal neurovascular coupling, response inhibition, or a clinical effect." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the two event-averaged peak latencies relative to event onset:\n1. theta_peak_latency_sec\n2. hbo_peak_latency_sec\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown, explanations, units, or extra fields.\n3. The JSON keys must be EXACTLY \"theta_peak_latency_sec\" and \"hbo_peak_latency_sec\".\n4. Both values must be finite JSON numbers or null if unavailable.\n5. Do not infer a missing latency from the interpretation.\n\n### OUTPUT TEMPLATE\n{\"theta_peak_latency_sec\": , \"hbo_peak_latency_sec\": }", + "metrics": [ + { + "metric_id": "theta_peak_latency_sec", + "type": "numeric_check", + "target_key": "theta_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 0.2, + "tolerance": 0.2 + } + }, + { + "metric_id": "hbo_peak_latency_sec", + "type": "numeric_check", + "target_key": "hbo_peak_latency_sec", + "weight": 35, + "params": { + "gt_value": 9.200000000000006, + "tolerance": 0.5 + } + }, + { + "metric_id": "neurovascular_timing_interpretation", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the response correctly describes the event-averaged temporal ordering.\n\nGround truth:\n- theta_peak_latency_sec = 0.200000 seconds;\n- hbo_peak_latency_sec = 9.200000 seconds;\n- the HbO peak occurs later than the frontal theta peak.\n\nA correct response must:\n1. State or clearly imply that frontal theta peaks earlier;\n2. State or clearly imply that the prefrontal HbO response peaks later;\n3. Describe the result as an electrophysiological-to-hemodynamic temporal sequence or delayed hemodynamic response;\n4. Treat this as event-averaged temporal association rather than proof of causal neurovascular coupling;\n5. Avoid claiming definitive response inhibition, clinical dysfunction, or unsupported statistical significance;\n6. Not contradict the reported latency values.\n\nReturn exactly {\"status\": true} only when these requirements are met. Otherwise return {\"status\": false}. Output one JSON object only, with no Markdown or additional fields." + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_01.json b/physiological_integration/cases/case34/case34_01.json new file mode 100644 index 0000000000000000000000000000000000000000..8c95028b5a2087ddd5feda3db09f32e85d907beb --- /dev/null +++ b/physiological_integration/cases/case34/case34_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `2` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_02.json b/physiological_integration/cases/case34/case34_02.json new file mode 100644 index 0000000000000000000000000000000000000000..f69d5e306afff2b4c8cddb5af24c0669447607ab --- /dev/null +++ b/physiological_integration/cases/case34/case34_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `6` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_03.json b/physiological_integration/cases/case34/case34_03.json new file mode 100644 index 0000000000000000000000000000000000000000..7e4e106b058f3f7af4db06a9968a3272b985c60a --- /dev/null +++ b/physiological_integration/cases/case34/case34_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `19` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_04.json b/physiological_integration/cases/case34/case34_04.json new file mode 100644 index 0000000000000000000000000000000000000000..afea6fc917deed9806057c08e07791a5a4143e04 --- /dev/null +++ b/physiological_integration/cases/case34/case34_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `24` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_05.json b/physiological_integration/cases/case34/case34_05.json new file mode 100644 index 0000000000000000000000000000000000000000..f2e2b03070361dd4b43643d66f3468bc17a31a92 --- /dev/null +++ b/physiological_integration/cases/case34/case34_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `26` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_06.json b/physiological_integration/cases/case34/case34_06.json new file mode 100644 index 0000000000000000000000000000000000000000..2bd7edd76c2643e3082c1a9777b2ead05f2708b5 --- /dev/null +++ b/physiological_integration/cases/case34/case34_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `32` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_07.json b/physiological_integration/cases/case34/case34_07.json new file mode 100644 index 0000000000000000000000000000000000000000..9984b9c702063c85ce254c303ee7ef5328d554e4 --- /dev/null +++ b/physiological_integration/cases/case34/case34_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `39` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_08.json b/physiological_integration/cases/case34/case34_08.json new file mode 100644 index 0000000000000000000000000000000000000000..6c3e592b8c0d81ec8b3ec1de657f5014217a4060 --- /dev/null +++ b/physiological_integration/cases/case34/case34_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze DEAP trial `40` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_09.json b/physiological_integration/cases/case34/case34_09.json new file mode 100644 index 0000000000000000000000000000000000000000..d718b1bdbe77fe84810043cb9ec5c6709a38d8c5 --- /dev/null +++ b/physiological_integration/cases/case34/case34_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `6` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_10.json b/physiological_integration/cases/case34/case34_10.json new file mode 100644 index 0000000000000000000000000000000000000000..a58c77569065dec22bc2baacf4a67fa673bbb887 --- /dev/null +++ b/physiological_integration/cases/case34/case34_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `11` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_11.json b/physiological_integration/cases/case34/case34_11.json new file mode 100644 index 0000000000000000000000000000000000000000..44e83d2495ef55acc1f57d5c4b5c36dffc40449b --- /dev/null +++ b/physiological_integration/cases/case34/case34_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `13` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_12.json b/physiological_integration/cases/case34/case34_12.json new file mode 100644 index 0000000000000000000000000000000000000000..a944e7ddd7f6e2d9b8dcdd5a169d895a91a9b7c5 --- /dev/null +++ b/physiological_integration/cases/case34/case34_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `20` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_13.json b/physiological_integration/cases/case34/case34_13.json new file mode 100644 index 0000000000000000000000000000000000000000..95a90f4985fbfebe674dfd6a7e486e366618e1f8 --- /dev/null +++ b/physiological_integration/cases/case34/case34_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `21` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EMG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_14.json b/physiological_integration/cases/case34/case34_14.json new file mode 100644 index 0000000000000000000000000000000000000000..a575182d03b17f162b7b03d791f175914148fc40 --- /dev/null +++ b/physiological_integration/cases/case34/case34_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `26` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_15.json b/physiological_integration/cases/case34/case34_15.json new file mode 100644 index 0000000000000000000000000000000000000000..8fe74e97330f74464a23b64659c6c791fa94d4e9 --- /dev/null +++ b/physiological_integration/cases/case34/case34_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `30` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_16.json b/physiological_integration/cases/case34/case34_16.json new file mode 100644 index 0000000000000000000000000000000000000000..998872baa4498a6ce53c04385d1bde2a62a8fab1 --- /dev/null +++ b/physiological_integration/cases/case34/case34_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze DEAP trial `36` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_17.json b/physiological_integration/cases/case34/case34_17.json new file mode 100644 index 0000000000000000000000000000000000000000..3efebeb991f45ace53ec31d2c9736ff3cadbc398 --- /dev/null +++ b/physiological_integration/cases/case34/case34_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `3` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_18.json b/physiological_integration/cases/case34/case34_18.json new file mode 100644 index 0000000000000000000000000000000000000000..bfe181de8f9b01ffe5f81bcf73ec202b32ce98f2 --- /dev/null +++ b/physiological_integration/cases/case34/case34_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `6` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_19.json b/physiological_integration/cases/case34/case34_19.json new file mode 100644 index 0000000000000000000000000000000000000000..438b312f9bfc691a21c0ba615fa27cb3d738f635 --- /dev/null +++ b/physiological_integration/cases/case34/case34_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `10` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_20.json b/physiological_integration/cases/case34/case34_20.json new file mode 100644 index 0000000000000000000000000000000000000000..1eb98bef4ed029a4eaa81f7286cf468823702e25 --- /dev/null +++ b/physiological_integration/cases/case34/case34_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `14` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_21.json b/physiological_integration/cases/case34/case34_21.json new file mode 100644 index 0000000000000000000000000000000000000000..f9fcdb424dfb810bd89c986811bbaada17197092 --- /dev/null +++ b/physiological_integration/cases/case34/case34_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `22` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_22.json b/physiological_integration/cases/case34/case34_22.json new file mode 100644 index 0000000000000000000000000000000000000000..e4f25d49e6339e2a0b88e200ae475b59f6e3c4c8 --- /dev/null +++ b/physiological_integration/cases/case34/case34_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `24` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_23.json b/physiological_integration/cases/case34/case34_23.json new file mode 100644 index 0000000000000000000000000000000000000000..a52e4fa6eccc9b915f59bf35018db5a64fba5d4e --- /dev/null +++ b/physiological_integration/cases/case34/case34_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `29` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_24.json b/physiological_integration/cases/case34/case34_24.json new file mode 100644 index 0000000000000000000000000000000000000000..87af4712b45a00521de1bfad6556fe0e0aa0425e --- /dev/null +++ b/physiological_integration/cases/case34/case34_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze DEAP trial `32` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EMG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_25.json b/physiological_integration/cases/case34/case34_25.json new file mode 100644 index 0000000000000000000000000000000000000000..fc1c84cad734b6f58a4af8da317a60a6db71c9f6 --- /dev/null +++ b/physiological_integration/cases/case34/case34_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `2` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_26.json b/physiological_integration/cases/case34/case34_26.json new file mode 100644 index 0000000000000000000000000000000000000000..a4b8ceca997ac7b404bc405d58bb18a60b380f0c --- /dev/null +++ b/physiological_integration/cases/case34/case34_26.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `15` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_27.json b/physiological_integration/cases/case34/case34_27.json new file mode 100644 index 0000000000000000000000000000000000000000..429b1866b0550877df2d164d12ea6c0c3362d0b6 --- /dev/null +++ b/physiological_integration/cases/case34/case34_27.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `20` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_28.json b/physiological_integration/cases/case34/case34_28.json new file mode 100644 index 0000000000000000000000000000000000000000..a0fb0710e05769c2616957eef0952f9a9ea3e52a --- /dev/null +++ b/physiological_integration/cases/case34/case34_28.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `24` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_29.json b/physiological_integration/cases/case34/case34_29.json new file mode 100644 index 0000000000000000000000000000000000000000..3ca7b130fb059c619d878a4364f6b118774bce81 --- /dev/null +++ b/physiological_integration/cases/case34/case34_29.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `28` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_30.json b/physiological_integration/cases/case34/case34_30.json new file mode 100644 index 0000000000000000000000000000000000000000..61b6ae474adffec2e2856f4c123059569a9599db --- /dev/null +++ b/physiological_integration/cases/case34/case34_30.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `30` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_31.json b/physiological_integration/cases/case34/case34_31.json new file mode 100644 index 0000000000000000000000000000000000000000..75a3227820f7603425dd1d8d6e9da9e1323d3c23 --- /dev/null +++ b/physiological_integration/cases/case34/case34_31.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `38` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_32.json b/physiological_integration/cases/case34/case34_32.json new file mode 100644 index 0000000000000000000000000000000000000000..1e719e2e7b22e1c8b35d63aede455a22605f09f6 --- /dev/null +++ b/physiological_integration/cases/case34/case34_32.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze DEAP trial `39` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_33.json b/physiological_integration/cases/case34/case34_33.json new file mode 100644 index 0000000000000000000000000000000000000000..b53b329f35488f548e0b5775f850806fcfb7d138 --- /dev/null +++ b/physiological_integration/cases/case34/case34_33.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `4` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EOG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_34.json b/physiological_integration/cases/case34/case34_34.json new file mode 100644 index 0000000000000000000000000000000000000000..a632a588a775d5cbcbd64fbaa17141c9cf84c42b --- /dev/null +++ b/physiological_integration/cases/case34/case34_34.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `18` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_35.json b/physiological_integration/cases/case34/case34_35.json new file mode 100644 index 0000000000000000000000000000000000000000..4a9cbc044ebe26972d2fca424b5692f73e5a384d --- /dev/null +++ b/physiological_integration/cases/case34/case34_35.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `24` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EMG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_36.json b/physiological_integration/cases/case34/case34_36.json new file mode 100644 index 0000000000000000000000000000000000000000..4a6b97fc285be547949b0016b3fcad4d745303ef --- /dev/null +++ b/physiological_integration/cases/case34/case34_36.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `26` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_37.json b/physiological_integration/cases/case34/case34_37.json new file mode 100644 index 0000000000000000000000000000000000000000..8907e0ffccc7430ec18ced6943b15cf43c154379 --- /dev/null +++ b/physiological_integration/cases/case34/case34_37.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `32` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_38.json b/physiological_integration/cases/case34/case34_38.json new file mode 100644 index 0000000000000000000000000000000000000000..0142a7e6c1f60a41cb085a3b5ca7356b1123cd83 --- /dev/null +++ b/physiological_integration/cases/case34/case34_38.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `36` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "A" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EMG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_39.json b/physiological_integration/cases/case34/case34_39.json new file mode 100644 index 0000000000000000000000000000000000000000..145b6347d83824ee4700912d43ef2d69e253f3dd --- /dev/null +++ b/physiological_integration/cases/case34/case34_39.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `38` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "C" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "EMG" + } + } + ] + } +} diff --git a/physiological_integration/cases/case34/case34_40.json b/physiological_integration/cases/case34/case34_40.json new file mode 100644 index 0000000000000000000000000000000000000000..3254e7fdf5bae1c082ad168619397a74a1c079e0 --- /dev/null +++ b/physiological_integration/cases/case34/case34_40.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case34_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze DEAP trial `39` and compare the trial-relative candidate intervals:\n\n- `A`: `[8, 12)`\n- `B`: `[24, 28)`\n- `C`: `[40, 44)`\n\nFor each candidate interval, use the immediately preceding and following\n4-second intervals as references. Combine the two reference intervals and use\ntheir median sample value as one common reference center. For every channel in\nthe following three modality groups, measure the root-mean-square deviation\nfrom this center in the candidate and reference intervals, normalize the\ncandidate disturbance by the average disturbance of the two references, and\naggregate channel scores using their median:\n\n- frontal EEG: `Fp1`, `Fp2`, `AF3`, `AF4`;\n- EOG: `EXG1`, `EXG2`;\n- EMG: `EXG3`, `EXG4`.\n\nUse the aggregated EEG score to select the candidate with the largest frontal\nEEG disturbance. Treat the disturbance as clear only when its score is at\nleast `1.25` and at least `20%` larger than the second-highest EEG score.\n\nFor the selected candidate, compare the aggregated EOG and EMG support scores.\nReport `EOG` only when the EOG support score is at least `1.25` and at least\n`10%` larger than the EMG score. Report `EMG` under the analogous condition.\nOtherwise report `neither`. If the EEG disturbance is not clear, report\n`neither` regardless of the EOG and EMG scores.\n\nReport only `selected_candidate` and `supporting_modality`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine.\n\n### TASK\nExtract:\n- `selected_candidate`: the selected candidate interval label.\n- `supporting_modality`: the modality supporting the frontal EEG disturbance.\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Do not output Markdown, explanations, intervals, or extra fields.\n3. `selected_candidate` must be exactly `A`, `B`, `C`, or null.\n4. `supporting_modality` must be exactly `EOG`, `EMG`, `neither`, or null.\n5. Do not infer or reconstruct an interval when only a candidate label is reported.\n\n### OUTPUT TEMPLATE\n{\"selected_candidate\":\"\",\"supporting_modality\":\"\"}", + "metrics": [ + { + "metric_id": "selected_candidate", + "type": "categorical_check", + "target_key": "selected_candidate", + "weight": 50, + "params": { + "gt_value": "B" + } + }, + { + "metric_id": "supporting_modality", + "type": "categorical_check", + "target_key": "supporting_modality", + "weight": 50, + "params": { + "gt_value": "neither" + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_01.json b/physiological_integration/cases/case35/case35_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d38afc95b7b209c7fc7fa7d02641c20aef0cf49e --- /dev/null +++ b/physiological_integration/cases/case35/case35_01.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[205.523, 268.842)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 3.4389125639305484e-05, + "tolerance": 6.877825127861097e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 1.59335016799248e-05, + "tolerance": 3.1867003359849604e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_02.json b/physiological_integration/cases/case35/case35_02.json new file mode 100644 index 0000000000000000000000000000000000000000..4fb2e7462127b441034c55d3f14f6c5fafad4338 --- /dev/null +++ b/physiological_integration/cases/case35/case35_02.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[78.629, 143.184)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00014810572729792863, + "tolerance": 2.9621145459585728e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -4.00234087006518e-05, + "tolerance": 8.00468174013036e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_03.json b/physiological_integration/cases/case35/case35_03.json new file mode 100644 index 0000000000000000000000000000000000000000..32200b2ea57c6ed493ab81fd0e7a52d7562d0cda --- /dev/null +++ b/physiological_integration/cases/case35/case35_03.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[131.246, 190.280)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 9.76254108657169e-06, + "tolerance": 1.952508217314338e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -4.642250468745831e-05, + "tolerance": 9.284500937491664e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_04.json b/physiological_integration/cases/case35/case35_04.json new file mode 100644 index 0000000000000000000000000000000000000000..ce844ff40b0e7fe58ddbf49f275c7c0a752184f3 --- /dev/null +++ b/physiological_integration/cases/case35/case35_04.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[142.132, 204.471)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 4.8291012554146886e-05, + "tolerance": 9.658202510829378e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -4.5884878521970276e-05, + "tolerance": 9.176975704394056e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_05.json b/physiological_integration/cases/case35/case35_05.json new file mode 100644 index 0000000000000000000000000000000000000000..4b21f68f6c3919bd7a770cc5e10849fbf9475de7 --- /dev/null +++ b/physiological_integration/cases/case35/case35_05.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[139.391, 205.806)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00010915956069161147, + "tolerance": 2.1831912138322296e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -5.746897258982386e-05, + "tolerance": 1.1493794517964773e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_06.json b/physiological_integration/cases/case35/case35_06.json new file mode 100644 index 0000000000000000000000000000000000000000..6da442a10f3c3d228d02e496b89db213ba5100cb --- /dev/null +++ b/physiological_integration/cases/case35/case35_06.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[254.431, 293.211)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.0002306962292359217, + "tolerance": 4.6139245847184346e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 5.5225082330723674e-05, + "tolerance": 1.1045016466144735e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_07.json b/physiological_integration/cases/case35/case35_07.json new file mode 100644 index 0000000000000000000000000000000000000000..de6f7d6dfe3809eca034d46463e8e848c39e84dd --- /dev/null +++ b/physiological_integration/cases/case35/case35_07.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[81.935, 147.614)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00014810572729792863, + "tolerance": 2.9621145459585728e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -4.00234087006518e-05, + "tolerance": 8.00468174013036e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_08.json b/physiological_integration/cases/case35/case35_08.json new file mode 100644 index 0000000000000000000000000000000000000000..62f7a10fa163f48dcda4f85371bcade18d18f5f1 --- /dev/null +++ b/physiological_integration/cases/case35/case35_08.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[133.391, 192.441)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -1.4064850938520803e-05, + "tolerance": 2.812970187704161e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -3.5501059163548875e-05, + "tolerance": 7.1002118327097754e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_09.json b/physiological_integration/cases/case35/case35_09.json new file mode 100644 index 0000000000000000000000000000000000000000..fd9199372997a06e6916206cfd310c7eaae71e70 --- /dev/null +++ b/physiological_integration/cases/case35/case35_09.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[146.416, 206.644)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 7.954960295070059e-05, + "tolerance": 1.590992059014012e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -5.9396948257460365e-05, + "tolerance": 1.1879389651492074e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_10.json b/physiological_integration/cases/case35/case35_10.json new file mode 100644 index 0000000000000000000000000000000000000000..d180ddc2422f1e615f7adbf5d384fdebff6a3863 --- /dev/null +++ b/physiological_integration/cases/case35/case35_10.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[142.598, 209.035)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00011421020305974773, + "tolerance": 2.284204061194955e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -7.126810020880718e-05, + "tolerance": 1.4253620041761436e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_11.json b/physiological_integration/cases/case35/case35_11.json new file mode 100644 index 0000000000000000000000000000000000000000..8a60bdea2a97ae2ed9304c0eac397b2c2935e5f1 --- /dev/null +++ b/physiological_integration/cases/case35/case35_11.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[256.658, 308.800)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.000271610561491928, + "tolerance": 5.432211229838561e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 7.52414982225277e-05, + "tolerance": 1.504829964450554e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_12.json b/physiological_integration/cases/case35/case35_12.json new file mode 100644 index 0000000000000000000000000000000000000000..f1692e3efe1815329bf8eb318ce3278c601fe94b --- /dev/null +++ b/physiological_integration/cases/case35/case35_12.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[85.235, 149.847)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00014936637652933636, + "tolerance": 2.9873275305867272e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -1.5120928758258046e-05, + "tolerance": 3.024185751651609e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_13.json b/physiological_integration/cases/case35/case35_13.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7a4c7537091c5d616ac68643b7196e8fd91881 --- /dev/null +++ b/physiological_integration/cases/case35/case35_13.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[135.543, 195.676)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -4.601269139796263e-05, + "tolerance": 9.202538279592528e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -2.1394727907922183e-05, + "tolerance": 4.2789455815844365e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_14.json b/physiological_integration/cases/case35/case35_14.json new file mode 100644 index 0000000000000000000000000000000000000000..ff0586cb8479932d42df9f552b51d336109fb4d4 --- /dev/null +++ b/physiological_integration/cases/case35/case35_14.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[149.637, 208.823)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 7.954960295070059e-05, + "tolerance": 1.590992059014012e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -5.9396948257460365e-05, + "tolerance": 1.1879389651492074e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_15.json b/physiological_integration/cases/case35/case35_15.json new file mode 100644 index 0000000000000000000000000000000000000000..9450cc82380ddb3dcc236550a33380c0750fb3fb --- /dev/null +++ b/physiological_integration/cases/case35/case35_15.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[145.823, 211.169)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00016685451765915805, + "tolerance": 3.337090353183161e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -8.161232824802585e-05, + "tolerance": 1.6322465649605172e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_16.json b/physiological_integration/cases/case35/case35_16.json new file mode 100644 index 0000000000000000000000000000000000000000..76b3efbcd922c936b96d486c2241eaf6f49e18b8 --- /dev/null +++ b/physiological_integration/cases/case35/case35_16.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[258.888, 323.294)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.00033378868697985864, + "tolerance": 6.675773739597173e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 8.047816698759575e-05, + "tolerance": 1.609563339751915e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_17.json b/physiological_integration/cases/case35/case35_17.json new file mode 100644 index 0000000000000000000000000000000000000000..9e9ba02fde1242edadf11554a7b99d3fc5f52e08 --- /dev/null +++ b/physiological_integration/cases/case35/case35_17.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[125.317, 186.692)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.0003036912054992374, + "tolerance": 6.0738241099847484e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 2.0957442583769136e-05, + "tolerance": 4.191488516753827e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_18.json b/physiological_integration/cases/case35/case35_18.json new file mode 100644 index 0000000000000000000000000000000000000000..22c34f059a9541140707248c332f3b655bbeb6b1 --- /dev/null +++ b/physiological_integration/cases/case35/case35_18.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[137.714, 198.904)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -4.922273600121467e-05, + "tolerance": 9.844547200242935e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -1.4569930791883907e-05, + "tolerance": 2.913986158376782e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_19.json b/physiological_integration/cases/case35/case35_19.json new file mode 100644 index 0000000000000000000000000000000000000000..b5bf48eae0ed5335a1a5dbda39fd15f8dba474c6 --- /dev/null +++ b/physiological_integration/cases/case35/case35_19.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[314.043, 353.030)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 8.137755119572055e-06, + "tolerance": 1.6275510239144112e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -2.457016018034642e-05, + "tolerance": 4.914032036069284e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_20.json b/physiological_integration/cases/case35/case35_20.json new file mode 100644 index 0000000000000000000000000000000000000000..60b02709d39fb9e90a1646fd241dde5a2a04a8e6 --- /dev/null +++ b/physiological_integration/cases/case35/case35_20.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[149.062, 213.312)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00016170355214784458, + "tolerance": 3.234071042956892e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -8.276701254006832e-05, + "tolerance": 1.6553402508013667e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_21.json b/physiological_integration/cases/case35/case35_21.json new file mode 100644 index 0000000000000000000000000000000000000000..6c21f566f7d77a2f459d68a25aa272eca7b470b3 --- /dev/null +++ b/physiological_integration/cases/case35/case35_21.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[261.107, 325.524)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.0003621276483259105, + "tolerance": 7.24255296651821e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 7.116450489781011e-05, + "tolerance": 1.4232900979562023e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_22.json b/physiological_integration/cases/case35/case35_22.json new file mode 100644 index 0000000000000000000000000000000000000000..33110b1c06b21d35a3eb10fa063c7eb84dd585d0 --- /dev/null +++ b/physiological_integration/cases/case35/case35_22.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[137.635, 201.077)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.0003008272049995564, + "tolerance": 6.0165440999911275e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -1.047262420097503e-05, + "tolerance": 2.0945248401950064e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_23.json b/physiological_integration/cases/case35/case35_23.json new file mode 100644 index 0000000000000000000000000000000000000000..201ee6394c37d18360021e78236952a787e6d69d --- /dev/null +++ b/physiological_integration/cases/case35/case35_23.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[140.953, 201.084)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -6.866895373220534e-05, + "tolerance": 1.373379074644107e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -9.61506645149399e-06, + "tolerance": 1.923013290298798e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_24.json b/physiological_integration/cases/case35/case35_24.json new file mode 100644 index 0000000000000000000000000000000000000000..b0721c0f2e5018b4bc23beada56236630124e3b3 --- /dev/null +++ b/physiological_integration/cases/case35/case35_24.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[335.703, 401.412)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 6.146306368510674e-05, + "tolerance": 1.229261273702135e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -9.913667236535158e-06, + "tolerance": 1.9827334473070316e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_25.json b/physiological_integration/cases/case35/case35_25.json new file mode 100644 index 0000000000000000000000000000000000000000..4462d95e069eda47df2d5c3e04bf08e5283edd25 --- /dev/null +++ b/physiological_integration/cases/case35/case35_25.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[152.315, 216.543)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00014042025680505762, + "tolerance": 2.8084051361011524e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -7.359217212524999e-05, + "tolerance": 1.4718434425049999e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_26.json b/physiological_integration/cases/case35/case35_26.json new file mode 100644 index 0000000000000000000000000000000000000000..3efed5f623668e4cea80ed8e6c865913ca18218b --- /dev/null +++ b/physiological_integration/cases/case35/case35_26.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[263.322, 328.855)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.00033172287913798843, + "tolerance": 6.63445758275977e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 5.4154647800841774e-05, + "tolerance": 1.0830929560168356e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_27.json b/physiological_integration/cases/case35/case35_27.json new file mode 100644 index 0000000000000000000000000000000000000000..b9e2d22fd740c58544fc1842c614fc4110ab3bea --- /dev/null +++ b/physiological_integration/cases/case35/case35_27.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[269.866, 341.056)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 1.9260675175539587e-05, + "tolerance": 3.852135035107918e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -2.9272026261529566e-05, + "tolerance": 5.854405252305914e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_28.json b/physiological_integration/cases/case35/case35_28.json new file mode 100644 index 0000000000000000000000000000000000000000..311172cb42f6a4841cf0f9522cd3c455ec0e1cc2 --- /dev/null +++ b/physiological_integration/cases/case35/case35_28.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[203.243, 264.709)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": -0.00016765761177217476, + "tolerance": 3.3531522354434956e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": 5.3087290232789694e-05, + "tolerance": 1.061745804655794e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_29.json b/physiological_integration/cases/case35/case35_29.json new file mode 100644 index 0000000000000000000000000000000000000000..2b3279636d4be0438a4a37ffdb7a0a84d40e3a82 --- /dev/null +++ b/physiological_integration/cases/case35/case35_29.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[337.856, 403.578)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 1.293630724858607e-05, + "tolerance": 2.587261449717214e-06 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -9.99338228182918e-06, + "tolerance": 1.9986764563658362e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case35/case35_30.json b/physiological_integration/cases/case35/case35_30.json new file mode 100644 index 0000000000000000000000000000000000000000..9b737c3c9a7f535a83ef71694511ed231fa70223 --- /dev/null +++ b/physiological_integration/cases/case35/case35_30.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "case35_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "label_path": "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "instruction": "Use all complete NoGo events (`event_code = 32`) whose onset falls within `[154.480, 219.784)` seconds.\n\nFor each event, compute theta power `[4, 8)` Hz over `FP1`, `AFz`, and `F1` for baseline `[-1, 0)` and response `[0, 1)` using a 1-second Hann-window Welch spectrum. Average the three channel powers and retain the event when response theta is higher than baseline theta.\n\nFor each retained event, use fNIRS channels `S03D02_760` and `S03D02_850`, with `[-2, 0)` as baseline and `[4, 8)` as response. Convert the `760`- and `850`-nm signals to HbO and HbR changes using the modified Beer–Lambert law with extinction coefficients `(1.4666, 3.8437)` and `(2.5264, 1.7986)`, a 3-cm source–detector distance, and DPF `6`. Use the mean baseline intensity as the optical-density reference and the natural logarithm for the intensity-to-optical-density conversion. Apply no additional fNIRS filtering and keep the resulting concentrations in mM.\n\nFor HbO and HbR separately, compute the response mean minus the baseline mean and average the changes across retained events.\n\nReport only `mean_hbo_change` and `mean_hbr_change`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for simultaneous EEG-fNIRS reports. Your sole task is to convert the agent's report into one JSON object.\n\n### TASK\nExtract the mean baseline-corrected responses across the theta-positive NoGo events:\n1. mean_hbo_change\n2. mean_hbr_change\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY one valid JSON object.\n2. DO NOT include Markdown code blocks, explanations, units, event counts, or extra fields.\n3. The JSON keys must be EXACTLY \"mean_hbo_change\" and \"mean_hbr_change\".\n4. Both values must be finite JSON numbers, or null when the requested value is missing or unclear.\n5. Do not return values from an individual event or an intermediate calculation.\n\n### OUTPUT TEMPLATE\n{\"mean_hbo_change\": , \"mean_hbr_change\": }", + "metrics": [ + { + "metric_id": "mean_hbo_change", + "type": "numeric_check", + "target_key": "mean_hbo_change", + "weight": 50, + "params": { + "gt_value": 0.00014951033774655877, + "tolerance": 2.9902067549311755e-05 + } + }, + { + "metric_id": "mean_hbr_change", + "type": "numeric_check", + "target_key": "mean_hbr_change", + "weight": 50, + "params": { + "gt_value": -6.0182891789202354e-05, + "tolerance": 1.2036578357840472e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_01.json b/physiological_integration/cases/case36/case36_01.json new file mode 100644 index 0000000000000000000000000000000000000000..a0671b06f3a41fcd5ee92d341a8791ea61144a62 --- /dev/null +++ b/physiological_integration/cases/case36/case36_01.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with both dominance and liking ratings at least `6`, regardless of valence or arousal.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.13636363636363635, + "tolerance": 0.022727272727272728 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_02.json b/physiological_integration/cases/case36/case36_02.json new file mode 100644 index 0000000000000000000000000000000000000000..fc8a80b2631494bc8bbb6a0806006fdb8e6ad22c --- /dev/null +++ b/physiological_integration/cases/case36/case36_02.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with both arousal and dominance ratings at least `6`, regardless of valence or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.15, + "tolerance": 0.025 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_03.json b/physiological_integration/cases/case36/case36_03.json new file mode 100644 index 0000000000000000000000000000000000000000..83d5d589be7ed9380fd505a75c1bbc2ce5a5af97 --- /dev/null +++ b/physiological_integration/cases/case36/case36_03.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with valence ratings at least `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.1111111111111111, + "tolerance": 0.027777777777777776 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_04.json b/physiological_integration/cases/case36/case36_04.json new file mode 100644 index 0000000000000000000000000000000000000000..98bbd1fd823744c383109b6a34b91e24117c5c96 --- /dev/null +++ b/physiological_integration/cases/case36/case36_04.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with both valence and liking ratings at least `6`, regardless of arousal or dominance.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.11764705882352941, + "tolerance": 0.029411764705882353 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_05.json b/physiological_integration/cases/case36/case36_05.json new file mode 100644 index 0000000000000000000000000000000000000000..10d96f3d70ee61bf74715c35f7bd2f495f0db7c7 --- /dev/null +++ b/physiological_integration/cases/case36/case36_05.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with dominance ratings at least `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.19230769230769232, + "tolerance": 0.019230769230769232 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_06.json b/physiological_integration/cases/case36/case36_06.json new file mode 100644 index 0000000000000000000000000000000000000000..28bcedf71b56313755a8fd79c1ed8f2df8c11e4a --- /dev/null +++ b/physiological_integration/cases/case36/case36_06.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with both valence and arousal ratings at least `6`, regardless of dominance or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.041666666666666664 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_07.json b/physiological_integration/cases/case36/case36_07.json new file mode 100644 index 0000000000000000000000000000000000000000..8320ef2e9cc20b786228e5e8f930b3cebeaddd09 --- /dev/null +++ b/physiological_integration/cases/case36/case36_07.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Select all DEAP trials with liking ratings at least `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.1935483870967742, + "tolerance": 0.016129032258064516 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_08.json b/physiological_integration/cases/case36/case36_08.json new file mode 100644 index 0000000000000000000000000000000000000000..431b52bc0a0904026a105cbbb18189795dad38df --- /dev/null +++ b/physiological_integration/cases/case36/case36_08.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Select all DEAP trials with arousal ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.16666666666666666, + "tolerance": 0.08333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_09.json b/physiological_integration/cases/case36/case36_09.json new file mode 100644 index 0000000000000000000000000000000000000000..3d97eec56b0cebd9c2f47c12f801bedb993fa9df --- /dev/null +++ b/physiological_integration/cases/case36/case36_09.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.02631578947368421 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_10.json b/physiological_integration/cases/case36/case36_10.json new file mode 100644 index 0000000000000000000000000000000000000000..f7e9fee69bd4f1ae5e7fa8698a57229e2d672b2a --- /dev/null +++ b/physiological_integration/cases/case36/case36_10.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with dominance ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.03333333333333333, + "tolerance": 0.016666666666666666 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_11.json b/physiological_integration/cases/case36/case36_11.json new file mode 100644 index 0000000000000000000000000000000000000000..7852a0b774c03d8fede7f91966c7b89498f56c92 --- /dev/null +++ b/physiological_integration/cases/case36/case36_11.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with arousal ratings at or below `4`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.041666666666666664, + "tolerance": 0.020833333333333332 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_12.json b/physiological_integration/cases/case36/case36_12.json new file mode 100644 index 0000000000000000000000000000000000000000..d7f8a4766b06843fc8f2c2f360c72612b2f62690 --- /dev/null +++ b/physiological_integration/cases/case36/case36_12.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with liking ratings at least `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.043478260869565216, + "tolerance": 0.021739130434782608 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_13.json b/physiological_integration/cases/case36/case36_13.json new file mode 100644 index 0000000000000000000000000000000000000000..90159f00e7efa59174008695abe6035888eb560d --- /dev/null +++ b/physiological_integration/cases/case36/case36_13.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6` and dominance ratings strictly between `4` and `6`, regardless of arousal or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.03125 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_14.json b/physiological_integration/cases/case36/case36_14.json new file mode 100644 index 0000000000000000000000000000000000000000..7e5617ad48a85ac252399ec6fe749d4588a408fb --- /dev/null +++ b/physiological_integration/cases/case36/case36_14.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with liking ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.03333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_15.json b/physiological_integration/cases/case36/case36_15.json new file mode 100644 index 0000000000000000000000000000000000000000..bdfe41785eeebc99c94dcc8243038c9452391f96 --- /dev/null +++ b/physiological_integration/cases/case36/case36_15.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with arousal ratings at or below `4` and liking ratings at least `6`, regardless of valence or dominance.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.058823529411764705, + "tolerance": 0.029411764705882353 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_16.json b/physiological_integration/cases/case36/case36_16.json new file mode 100644 index 0000000000000000000000000000000000000000..27fc0d08f0de1c6fdf25c32cfbbc0fcf78a244b2 --- /dev/null +++ b/physiological_integration/cases/case36/case36_16.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with dominance ratings strictly between `4` and `6` and liking ratings strictly between `4` and `6`, regardless of valence or arousal.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.038461538461538464 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_17.json b/physiological_integration/cases/case36/case36_17.json new file mode 100644 index 0000000000000000000000000000000000000000..f12d62bd44b109dbfdc902e5564d2a2b1a3c42ed --- /dev/null +++ b/physiological_integration/cases/case36/case36_17.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6` and liking ratings strictly between `4` and `6`, regardless of arousal or dominance.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.041666666666666664 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_18.json b/physiological_integration/cases/case36/case36_18.json new file mode 100644 index 0000000000000000000000000000000000000000..14dcc08a1c956d9cf45d38f1f497ee1035017740 --- /dev/null +++ b/physiological_integration/cases/case36/case36_18.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with dominance ratings strictly between `4` and `6` and liking ratings at least `6`, regardless of valence or arousal.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.06666666666666667, + "tolerance": 0.03333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_19.json b/physiological_integration/cases/case36/case36_19.json new file mode 100644 index 0000000000000000000000000000000000000000..fb11782b91740a13851be187b7939322262e3c34 --- /dev/null +++ b/physiological_integration/cases/case36/case36_19.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with dominance ratings at or below `4`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_20.json b/physiological_integration/cases/case36/case36_20.json new file mode 100644 index 0000000000000000000000000000000000000000..77e1834b243f1ef9d2bfe666317c2f7b1dd776b8 --- /dev/null +++ b/physiological_integration/cases/case36/case36_20.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with arousal ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.041666666666666664 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_21.json b/physiological_integration/cases/case36/case36_21.json new file mode 100644 index 0000000000000000000000000000000000000000..bea237f167a56c3ffff7bfaf8950f905e74383e9 --- /dev/null +++ b/physiological_integration/cases/case36/case36_21.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6` and arousal ratings at or below `4`, regardless of dominance or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.0625 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_22.json b/physiological_integration/cases/case36/case36_22.json new file mode 100644 index 0000000000000000000000000000000000000000..112bf7401ad5e66d058cad36b74aeab03d5d43e0 --- /dev/null +++ b/physiological_integration/cases/case36/case36_22.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6` and liking ratings at least `6`, regardless of arousal or dominance.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -2.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.08333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_23.json b/physiological_integration/cases/case36/case36_23.json new file mode 100644 index 0000000000000000000000000000000000000000..4430d396710e10c6f82465f7f96b1e18638473fd --- /dev/null +++ b/physiological_integration/cases/case36/case36_23.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings at least `6` and dominance ratings strictly between `4` and `6`, regardless of arousal or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.07692307692307693, + "tolerance": 0.038461538461538464 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_24.json b/physiological_integration/cases/case36/case36_24.json new file mode 100644 index 0000000000000000000000000000000000000000..008487b768775422924ec5d785ac7f49ebb7ffd8 --- /dev/null +++ b/physiological_integration/cases/case36/case36_24.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings at least `6` and arousal ratings at or below `4`, regardless of dominance or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -0.5, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.08333333333333333, + "tolerance": 0.041666666666666664 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_25.json b/physiological_integration/cases/case36/case36_25.json new file mode 100644 index 0000000000000000000000000000000000000000..f952ff26dfa1cc16525b876101d241582292ef3e --- /dev/null +++ b/physiological_integration/cases/case36/case36_25.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with dominance ratings at or below `4` and liking ratings at least `6`, regardless of valence or arousal.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.0625 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_26.json b/physiological_integration/cases/case36/case36_26.json new file mode 100644 index 0000000000000000000000000000000000000000..016884426af21b4b6a3010df960075c64528b07d --- /dev/null +++ b/physiological_integration/cases/case36/case36_26.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with arousal ratings at or below `4` and liking ratings strictly between `4` and `6`, regardless of valence or dominance.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.07142857142857142 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_27.json b/physiological_integration/cases/case36/case36_27.json new file mode 100644 index 0000000000000000000000000000000000000000..8c95ef332dd60aa69f435c9e8697dd4815be62aa --- /dev/null +++ b/physiological_integration/cases/case36/case36_27.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Select all DEAP trials with valence ratings strictly between `4` and `6` and arousal ratings strictly between `4` and `6`, regardless of dominance or liking.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.0, + "tolerance": 0.07142857142857142 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_28.json b/physiological_integration/cases/case36/case36_28.json new file mode 100644 index 0000000000000000000000000000000000000000..c69d5a24a79b44b62b434decdf95aa179c7f7d87 --- /dev/null +++ b/physiological_integration/cases/case36/case36_28.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Select all DEAP trials with valence ratings at least `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.15384615384615385, + "tolerance": 0.038461538461538464 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_29.json b/physiological_integration/cases/case36/case36_29.json new file mode 100644 index 0000000000000000000000000000000000000000..f5cc3154efa44b70413505cc72eb0dd5323c9e89 --- /dev/null +++ b/physiological_integration/cases/case36/case36_29.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Select all DEAP trials with arousal ratings strictly between `4` and `6`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.2, + "tolerance": 0.03333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case36/case36_30.json b/physiological_integration/cases/case36/case36_30.json new file mode 100644 index 0000000000000000000000000000000000000000..d6158a0ef53c6bbe76e7cbffcaddfbe807996a62 --- /dev/null +++ b/physiological_integration/cases/case36/case36_30.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "case36_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Select all DEAP trials with dominance ratings at or below `4`, regardless of the other ratings.\n\nMatch rating row `i` to trial `i` in chronological order, using `Status = 4` as stimulus onset and its following `Status = 5` marker as stimulus end.\n\nFor each selected trial, use the 3 seconds before onset as baseline and divide the following 60 seconds into six consecutive 10-second windows. Exclude a trial only if these intervals are not fully available.\n\nFor EEG channels `P3`, `P4`, `PO3`, `PO4`, `O1`, and `O2`, compute alpha power in `[8, 13)` Hz using a 2-second Hann-window Welch PSD with 50% overlap. Average alpha power across the six channels first, then compute each window's suppression as `log(baseline power + 1e-12) - log(window power + 1e-12)`. Select the window with the largest suppression.\n\nFor `GSR1`, subtract the baseline mean from each window mean and select the window with the largest increase. If maxima are tied, use the earliest window.\n\nFor each trial, define `peak lag = GSR peak-window index - EEG peak-window index`. Report the median lag across trials as `median_peak_lag_windows` and the fraction of trials with positive lag as `gsr_after_eeg_fraction`.\n\nReport only `median_peak_lag_windows` and `gsr_after_eeg_fraction`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for a DEAP cortical--autonomic timing analysis.\n\n### TASK\nExtract the median GSR-minus-EEG peak-window lag and the fraction of selected trials whose GSR peak occurs later than the EEG peak.\n\n### REQUIRED FIELDS\n- `median_peak_lag_windows`: a finite number, or `null`\n- `gsr_after_eeg_fraction`: a finite number between 0 and 1, or `null`\n\n### STRICT CONSTRAINTS\n1. Output only one valid JSON object.\n2. Use exactly the two required keys.\n3. Do not include Markdown, explanations, units, labels, or extra fields.\n4. Preserve the lag direction: positive values mean that the GSR peak is later.\n5. Do not calculate or infer a missing value.\n\n### OUTPUT TEMPLATE\n{\"median_peak_lag_windows\": , \"gsr_after_eeg_fraction\": }", + "metrics": [ + { + "metric_id": "median_peak_lag_windows", + "type": "numeric_check", + "target_key": "median_peak_lag_windows", + "weight": 50, + "params": { + "gt_value": -1.0, + "tolerance": 0.01 + } + }, + { + "metric_id": "gsr_after_eeg_fraction", + "type": "numeric_check", + "target_key": "gsr_after_eeg_fraction", + "weight": 50, + "params": { + "gt_value": 0.3333333333333333, + "tolerance": 0.027777777777777776 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_01.json b/physiological_integration/cases/case37/case37_01.json new file mode 100644 index 0000000000000000000000000000000000000000..067d36296144b29d7fe4655e3f5a51b78818eca9 --- /dev/null +++ b/physiological_integration/cases/case37/case37_01.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [870, 990)\n- `interval_B`: [930, 1050)\n- `interval_C`: [2490, 2610)\n- `interval_D`: [4290, 4410)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.883157780432186, + "tolerance": 0.0883157780432186 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_02.json b/physiological_integration/cases/case37/case37_02.json new file mode 100644 index 0000000000000000000000000000000000000000..724dc032c03481f7f264d7ab97abf10486e3b39a --- /dev/null +++ b/physiological_integration/cases/case37/case37_02.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [870, 990)\n- `interval_B`: [900, 1020)\n- `interval_C`: [2580, 2700)\n- `interval_D`: [4890, 5010)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "B", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8527287537333695, + "tolerance": 0.08527287537333696 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_03.json b/physiological_integration/cases/case37/case37_03.json new file mode 100644 index 0000000000000000000000000000000000000000..2882b9c9a2b60a6f981e46a74537cdde6990ebee --- /dev/null +++ b/physiological_integration/cases/case37/case37_03.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [690, 810)\n- `interval_B`: [870, 990)\n- `interval_C`: [3810, 3930)\n- `interval_D`: [5040, 5160)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "B", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "B", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8836300477637863, + "tolerance": 0.08836300477637864 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_04.json b/physiological_integration/cases/case37/case37_04.json new file mode 100644 index 0000000000000000000000000000000000000000..97cf8376d4d972837851d7c06f7dffdc2aa449e5 --- /dev/null +++ b/physiological_integration/cases/case37/case37_04.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [870, 990)\n- `interval_B`: [930, 1050)\n- `interval_C`: [2070, 2190)\n- `interval_D`: [3210, 3330)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8839179063959169, + "tolerance": 0.0883917906395917 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_05.json b/physiological_integration/cases/case37/case37_05.json new file mode 100644 index 0000000000000000000000000000000000000000..0f0f2a893583569626a1bb3485d62d8fe95af45d --- /dev/null +++ b/physiological_integration/cases/case37/case37_05.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [660, 780)\n- `interval_B`: [900, 1020)\n- `interval_C`: [3870, 3990)\n- `interval_D`: [4350, 4470)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.7992835432045159, + "tolerance": 0.07992835432045159 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_06.json b/physiological_integration/cases/case37/case37_06.json new file mode 100644 index 0000000000000000000000000000000000000000..38f7311f7366f646c751f45b3b05d226a69167b2 --- /dev/null +++ b/physiological_integration/cases/case37/case37_06.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_01_session-20151124_noon_2.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [630, 750)\n- `interval_B`: [870, 990)\n- `interval_C`: [3000, 3120)\n- `interval_D`: [4740, 4860)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "C", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "C", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8262049500651323, + "tolerance": 0.08262049500651324 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_07.json b/physiological_integration/cases/case37/case37_07.json new file mode 100644 index 0000000000000000000000000000000000000000..3dbd6c91689bef7a140e480fd748413bca085de5 --- /dev/null +++ b/physiological_integration/cases/case37/case37_07.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [240, 360)\n- `interval_B`: [300, 420)\n- `interval_C`: [3120, 3240)\n- `interval_D`: [5850, 5970)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "A", + "D", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "A", + "B", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.4834401129330004, + "tolerance": 0.04834401129330004 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_08.json b/physiological_integration/cases/case37/case37_08.json new file mode 100644 index 0000000000000000000000000000000000000000..91893d3b4f5005e7c69dd3d892c6876ee6709b27 --- /dev/null +++ b/physiological_integration/cases/case37/case37_08.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [270, 390)\n- `interval_B`: [450, 570)\n- `interval_C`: [5880, 6000)\n- `interval_D`: [6450, 6570)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.461029092488059, + "tolerance": 0.046102909248805904 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_09.json b/physiological_integration/cases/case37/case37_09.json new file mode 100644 index 0000000000000000000000000000000000000000..bfee90a08a0bf4371a67a1654acf40fef558a4c5 --- /dev/null +++ b/physiological_integration/cases/case37/case37_09.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [390, 510)\n- `interval_B`: [420, 540)\n- `interval_C`: [930, 1050)\n- `interval_D`: [5790, 5910)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "B", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.5475427640510454, + "tolerance": 0.054754276405104545 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_10.json b/physiological_integration/cases/case37/case37_10.json new file mode 100644 index 0000000000000000000000000000000000000000..a1d4ec0220057b44905d8cdc15390c4031e2ebd7 --- /dev/null +++ b/physiological_integration/cases/case37/case37_10.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [270, 390)\n- `interval_B`: [330, 450)\n- `interval_C`: [1950, 2070)\n- `interval_D`: [2940, 3060)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "A", + "B", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "C", + "A", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.4311000108589423, + "tolerance": 0.04311000108589423 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_11.json b/physiological_integration/cases/case37/case37_11.json new file mode 100644 index 0000000000000000000000000000000000000000..272dfeb148145468ae40a610f23c9140e33de243 --- /dev/null +++ b/physiological_integration/cases/case37/case37_11.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [270, 390)\n- `interval_B`: [330, 450)\n- `interval_C`: [1950, 2070)\n- `interval_D`: [3120, 3240)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "A", + "B", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "C", + "A", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.4903898360299707, + "tolerance": 0.04903898360299708 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_12.json b/physiological_integration/cases/case37/case37_12.json new file mode 100644 index 0000000000000000000000000000000000000000..4689d93355c73c90ca2afdd30f7eced01f3e8cce --- /dev/null +++ b/physiological_integration/cases/case37/case37_12.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_02_session-20151106_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [150, 270)\n- `interval_B`: [210, 330)\n- `interval_C`: [570, 690)\n- `interval_D`: [3120, 3240)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "D", + "B", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "C", + "B", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.3786513193614942, + "tolerance": 0.037865131936149425 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_13.json b/physiological_integration/cases/case37/case37_13.json new file mode 100644 index 0000000000000000000000000000000000000000..09b842b0fbe23ff61d204b55007a7f713b6bf5ad --- /dev/null +++ b/physiological_integration/cases/case37/case37_13.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [3510, 3630)\n- `interval_B`: [3600, 3720)\n- `interval_C`: [3900, 4020)\n- `interval_D`: [5460, 5580)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "A", + "B", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "D", + "C", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8162966179903371, + "tolerance": 0.08162966179903372 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_14.json b/physiological_integration/cases/case37/case37_14.json new file mode 100644 index 0000000000000000000000000000000000000000..6a0ad11c553b1a3900bb7e84dc4c7fcfbe9b71bc --- /dev/null +++ b/physiological_integration/cases/case37/case37_14.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [3570, 3690)\n- `interval_B`: [3600, 3720)\n- `interval_C`: [6240, 6360)\n- `interval_D`: [6540, 6660)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "C", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6211783871843606, + "tolerance": 0.06211783871843606 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_15.json b/physiological_integration/cases/case37/case37_15.json new file mode 100644 index 0000000000000000000000000000000000000000..cf927c39d9463d1dbf7cd17a1dba461332baa94c --- /dev/null +++ b/physiological_integration/cases/case37/case37_15.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [2700, 2820)\n- `interval_B`: [2910, 3030)\n- `interval_C`: [3300, 3420)\n- `interval_D`: [3810, 3930)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.7785497177594441, + "tolerance": 0.07785497177594441 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_16.json b/physiological_integration/cases/case37/case37_16.json new file mode 100644 index 0000000000000000000000000000000000000000..8cce49d10621e92a79a7ceb03227df16d545e1d5 --- /dev/null +++ b/physiological_integration/cases/case37/case37_16.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [2160, 2280)\n- `interval_B`: [2370, 2490)\n- `interval_C`: [3090, 3210)\n- `interval_D`: [3900, 4020)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "D", + "C", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8013460703430307, + "tolerance": 0.08013460703430308 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_17.json b/physiological_integration/cases/case37/case37_17.json new file mode 100644 index 0000000000000000000000000000000000000000..595345bd20b9b810574cfaca2febb70515cbdd4b --- /dev/null +++ b/physiological_integration/cases/case37/case37_17.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [2100, 2220)\n- `interval_B`: [2340, 2460)\n- `interval_C`: [2640, 2760)\n- `interval_D`: [3360, 3480)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "B", + "D", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "B", + "D", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.7813721233174119, + "tolerance": 0.0781372123317412 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_18.json b/physiological_integration/cases/case37/case37_18.json new file mode 100644 index 0000000000000000000000000000000000000000..00ca4add95120518646a866e53e3f38b27c5e93d --- /dev/null +++ b/physiological_integration/cases/case37/case37_18.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_03_session-20151024_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [3450, 3570)\n- `interval_B`: [3540, 3660)\n- `interval_C`: [5400, 5520)\n- `interval_D`: [6300, 6420)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "B", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8046794419412626, + "tolerance": 0.08046794419412627 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_19.json b/physiological_integration/cases/case37/case37_19.json new file mode 100644 index 0000000000000000000000000000000000000000..c6d085df77dec859325e72da9948350c9b496c92 --- /dev/null +++ b/physiological_integration/cases/case37/case37_19.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [4110, 4230)\n- `interval_B`: [4170, 4290)\n- `interval_C`: [6000, 6120)\n- `interval_D`: [6540, 6660)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "C", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6398088826148333, + "tolerance": 0.06398088826148333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_20.json b/physiological_integration/cases/case37/case37_20.json new file mode 100644 index 0000000000000000000000000000000000000000..ea23b2b3195098a4cccc2292456c9eb48e50ab09 --- /dev/null +++ b/physiological_integration/cases/case37/case37_20.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [4140, 4260)\n- `interval_B`: [4200, 4320)\n- `interval_C`: [4950, 5070)\n- `interval_D`: [6540, 6660)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "C", + "B", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "C", + "B", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6186339450537518, + "tolerance": 0.06186339450537518 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_21.json b/physiological_integration/cases/case37/case37_21.json new file mode 100644 index 0000000000000000000000000000000000000000..1c9da4000d4323dc3db1ff3d34c6836387c1b4a2 --- /dev/null +++ b/physiological_integration/cases/case37/case37_21.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [3420, 3540)\n- `interval_B`: [3570, 3690)\n- `interval_C`: [6090, 6210)\n- `interval_D`: [6720, 6840)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "B", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.549174989144594, + "tolerance": 0.0549174989144594 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_22.json b/physiological_integration/cases/case37/case37_22.json new file mode 100644 index 0000000000000000000000000000000000000000..d2faf455c5fea4185f9e03dd3eab8f69cc188494 --- /dev/null +++ b/physiological_integration/cases/case37/case37_22.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [4140, 4260)\n- `interval_B`: [4320, 4440)\n- `interval_C`: [4830, 4950)\n- `interval_D`: [5220, 5340)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "D", + "A", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "A", + "C", + "D", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6328701693443335, + "tolerance": 0.06328701693443335 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_23.json b/physiological_integration/cases/case37/case37_23.json new file mode 100644 index 0000000000000000000000000000000000000000..fcc1a5c855605222dbab994623c478c14cb54d9e --- /dev/null +++ b/physiological_integration/cases/case37/case37_23.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [3090, 3210)\n- `interval_B`: [3270, 3390)\n- `interval_C`: [5190, 5310)\n- `interval_D`: [6720, 6840)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "D", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6183239253148068, + "tolerance": 0.061832392531480686 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_24.json b/physiological_integration/cases/case37/case37_24.json new file mode 100644 index 0000000000000000000000000000000000000000..a3cad22fa0ada24b583f2c96cb751f7451f72b46 --- /dev/null +++ b/physiological_integration/cases/case37/case37_24.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_04_session-20151105_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [2490, 2610)\n- `interval_B`: [2520, 2640)\n- `interval_C`: [5160, 5280)\n- `interval_D`: [6000, 6120)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "B", + "A", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "C", + "D" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.5000814553353244, + "tolerance": 0.050008145533532444 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_25.json b/physiological_integration/cases/case37/case37_25.json new file mode 100644 index 0000000000000000000000000000000000000000..471ae0d014b86532e36ff88cdab28ff5439b56c0 --- /dev/null +++ b/physiological_integration/cases/case37/case37_25.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [2730, 2850)\n- `interval_B`: [2760, 2880)\n- `interval_C`: [4320, 4440)\n- `interval_D`: [6660, 6780)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "A", + "D", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.8110236220472442, + "tolerance": 0.08110236220472443 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_26.json b/physiological_integration/cases/case37/case37_26.json new file mode 100644 index 0000000000000000000000000000000000000000..a43a2c3fbcb3c5a999714e2ee6ac94a145a27614 --- /dev/null +++ b/physiological_integration/cases/case37/case37_26.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [1380, 1500)\n- `interval_B`: [1620, 1740)\n- `interval_C`: [6000, 6120)\n- `interval_D`: [6660, 6780)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "D", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "C", + "B", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6234259661311333, + "tolerance": 0.06234259661311334 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_27.json b/physiological_integration/cases/case37/case37_27.json new file mode 100644 index 0000000000000000000000000000000000000000..771a0b888d0cc2af9a20134c9ea647c9920ae95a --- /dev/null +++ b/physiological_integration/cases/case37/case37_27.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [1980, 2100)\n- `interval_B`: [2160, 2280)\n- `interval_C`: [4140, 4260)\n- `interval_D`: [6420, 6540)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "B", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "B", + "A", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.6172383847155885, + "tolerance": 0.06172383847155885 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_28.json b/physiological_integration/cases/case37/case37_28.json new file mode 100644 index 0000000000000000000000000000000000000000..394f8661a9be7de38c41bbe37f5f5463b2a91bc1 --- /dev/null +++ b/physiological_integration/cases/case37/case37_28.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [810, 930)\n- `interval_B`: [840, 960)\n- `interval_C`: [3930, 4050)\n- `interval_D`: [6210, 6330)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "A", + "B", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "A", + "B", + "C" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.7646483844691827, + "tolerance": 0.07646483844691827 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_29.json b/physiological_integration/cases/case37/case37_29.json new file mode 100644 index 0000000000000000000000000000000000000000..8e1ffed60c1d478d86aa5ce27b2fd84fe148c465 --- /dev/null +++ b/physiological_integration/cases/case37/case37_29.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [1530, 1650)\n- `interval_B`: [1620, 1740)\n- `interval_C`: [6030, 6150)\n- `interval_D`: [6480, 6600)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "B", + "C", + "D", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "D", + "C", + "B", + "A" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.5945931274089354, + "tolerance": 0.05945931274089355 + } + } + ] + } +} diff --git a/physiological_integration/cases/case37/case37_30.json b/physiological_integration/cases/case37/case37_30.json new file mode 100644 index 0000000000000000000000000000000000000000..de0f7880c78dcc1ebb77cf4712b0a2809b041322 --- /dev/null +++ b/physiological_integration/cases/case37/case37_30.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "case37_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVIG_05_session-20141108_noon.edf" + ], + "instruction": "Compare four 120-second intervals:\n\n- `interval_A`: [810, 930)\n- `interval_B`: [930, 1050)\n- `interval_C`: [2850, 2970)\n- `interval_D`: [4170, 4290)\n\nDivide each interval into consecutive 10-second epochs.\n\nFor each epoch, compute EEG slowing from `T7`, `T8`, `P1`, `P2`, `O1`, and `O2`. Average each band power across the six channels, then compute `log((theta + alpha) / (beta + gamma))`.\n\nFor `EOG-H` and `EOG-V`, average the `[0.1, 1.0)` Hz powers across the two channels and separately average the `[1.0, 10.0)` Hz powers, then compute `log(P_slow / P_fast)`.\n\nAverage the epoch-level EEG and EOG features within each interval and rank the four intervals from strongest to weakest evidence. Using all 48 paired epoch values, compute the Spearman correlation between EEG and EOG evidence.\n\nReport `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for EEG-EOG vigilance reports.\n\n### TASK\nExtract:\n- eeg_drowsiness_order: the four intervals ranked from strongest to weakest EEG drowsiness evidence.\n- eog_drowsiness_order: the four intervals ranked from strongest to weakest EOG drowsiness evidence.\n- eeg_eog_correlation: the Spearman correlation computed from all 48 paired epoch-level EEG and EOG values.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, units, or extra fields.\n3. The keys must be exactly `eeg_drowsiness_order`, `eog_drowsiness_order`, and `eeg_eog_correlation`.\n4. Each order must be an array of exactly four labels in strongest-to-weakest order and must contain `A`, `B`, `C`, and `D` exactly once.\n5. Normalize interval names before output: `interval_A` and `A` both mean `A`; `interval_B` and `B` both mean `B`; `interval_C` and `C` both mean `C`; `interval_D` and `D` both mean `D`. In particular, `interval_D` and `D` are the same label.\n6. Do not output the `interval_` prefix in either order.\n7. `eeg_eog_correlation` must be a finite number in [-1, 1], or null if unavailable.\n\n### OUTPUT TEMPLATE\n{\"eeg_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eog_drowsiness_order\": [\"\", \"\", \"\", \"\"], \"eeg_eog_correlation\": }", + "metrics": [ + { + "metric_id": "eeg_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eeg_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "A", + "D", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eog_drowsiness_order", + "type": "sequence_match_check", + "target_key": "eog_drowsiness_order", + "weight": 35, + "params": { + "gt_value": [ + "C", + "A", + "D", + "B" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "eeg_eog_correlation", + "type": "numeric_check", + "target_key": "eeg_eog_correlation", + "weight": 30, + "params": { + "gt_value": 0.7381676074685192, + "tolerance": 0.07381676074685192 + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_01.json b/physiological_integration/cases/case38/case38_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d8814ef9de996f330965bc7c868d0723f2d046ea --- /dev/null +++ b/physiological_integration/cases/case38/case38_01.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 15)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_02.json b/physiological_integration/cases/case38/case38_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b5f28ec228370411c9cb1dfe91005e83658af0c5 --- /dev/null +++ b/physiological_integration/cases/case38/case38_02.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_03.json b/physiological_integration/cases/case38/case38_03.json new file mode 100644 index 0000000000000000000000000000000000000000..62f48f54cbedd11e357bba78c941ae1990d009d0 --- /dev/null +++ b/physiological_integration/cases/case38/case38_03.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 45)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_04.json b/physiological_integration/cases/case38/case38_04.json new file mode 100644 index 0000000000000000000000000000000000000000..20cc217f08ff3012163aa53da0f9942d421d98af --- /dev/null +++ b/physiological_integration/cases/case38/case38_04.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 20)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_05.json b/physiological_integration/cases/case38/case38_05.json new file mode 100644 index 0000000000000000000000000000000000000000..1f1513e7bec7aa39e1fb0f6628f9dd0d2d9c4548 --- /dev/null +++ b/physiological_integration/cases/case38/case38_05.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[20, 40)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_06.json b/physiological_integration/cases/case38/case38_06.json new file mode 100644 index 0000000000000000000000000000000000000000..37366c9b076fb4178c5c4cef955d7d047cea46d5 --- /dev/null +++ b/physiological_integration/cases/case38/case38_06.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_07.json b/physiological_integration/cases/case38/case38_07.json new file mode 100644 index 0000000000000000000000000000000000000000..7aa0b43be81676e6f568650dabd94d5d2b8bb5d5 --- /dev/null +++ b/physiological_integration/cases/case38/case38_07.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[10, 50)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_08.json b/physiological_integration/cases/case38/case38_08.json new file mode 100644 index 0000000000000000000000000000000000000000..3cbe8d2b58123b182f2f86847e499103d011a9f4 --- /dev/null +++ b/physiological_integration/cases/case38/case38_08.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "label_path": "data/multi/DEAP_01.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_09.json b/physiological_integration/cases/case38/case38_09.json new file mode 100644 index 0000000000000000000000000000000000000000..14c7981899eda07eaace42c1c07fe1eaba6ee4eb --- /dev/null +++ b/physiological_integration/cases/case38/case38_09.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 15)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_10.json b/physiological_integration/cases/case38/case38_10.json new file mode 100644 index 0000000000000000000000000000000000000000..45a68c280b11ea5304c85f66f1ab63353028382c --- /dev/null +++ b/physiological_integration/cases/case38/case38_10.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_11.json b/physiological_integration/cases/case38/case38_11.json new file mode 100644 index 0000000000000000000000000000000000000000..f9911f74f2447b6288fddf8db99634e697fd4d60 --- /dev/null +++ b/physiological_integration/cases/case38/case38_11.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[45, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_12.json b/physiological_integration/cases/case38/case38_12.json new file mode 100644 index 0000000000000000000000000000000000000000..2f6e79d22385eec4d1d90cc7c268c7d3364746df --- /dev/null +++ b/physiological_integration/cases/case38/case38_12.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 20)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_13.json b/physiological_integration/cases/case38/case38_13.json new file mode 100644 index 0000000000000000000000000000000000000000..38308d7769fa694797b49ddcc946b047b6ac2ad9 --- /dev/null +++ b/physiological_integration/cases/case38/case38_13.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[20, 40)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_14.json b/physiological_integration/cases/case38/case38_14.json new file mode 100644 index 0000000000000000000000000000000000000000..0336b07f4b9aa4b49eef5780e527a0b02950842d --- /dev/null +++ b/physiological_integration/cases/case38/case38_14.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_15.json b/physiological_integration/cases/case38/case38_15.json new file mode 100644 index 0000000000000000000000000000000000000000..bd792539d91d16c3fe65522a845b2529f9a89044 --- /dev/null +++ b/physiological_integration/cases/case38/case38_15.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "dominance" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_16.json b/physiological_integration/cases/case38/case38_16.json new file mode 100644 index 0000000000000000000000000000000000000000..19469570034a468c6b5c981f32a1406166ae4192 --- /dev/null +++ b/physiological_integration/cases/case38/case38_16.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "label_path": "data/multi/DEAP_02.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[10, 50)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_17.json b/physiological_integration/cases/case38/case38_17.json new file mode 100644 index 0000000000000000000000000000000000000000..4e6c796595910f0fc6214ea38c89121e11871527 --- /dev/null +++ b/physiological_integration/cases/case38/case38_17.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 15)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_18.json b/physiological_integration/cases/case38/case38_18.json new file mode 100644 index 0000000000000000000000000000000000000000..1fab657ddb10436ec74f397269b3c9497f32b17d --- /dev/null +++ b/physiological_integration/cases/case38/case38_18.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_19.json b/physiological_integration/cases/case38/case38_19.json new file mode 100644 index 0000000000000000000000000000000000000000..59ca806a37777f136df9b749c47af641cbe322f7 --- /dev/null +++ b/physiological_integration/cases/case38/case38_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 45)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_20.json b/physiological_integration/cases/case38/case38_20.json new file mode 100644 index 0000000000000000000000000000000000000000..12aceebfe25598216579aaa5bde29c3cd53fef03 --- /dev/null +++ b/physiological_integration/cases/case38/case38_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[45, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_21.json b/physiological_integration/cases/case38/case38_21.json new file mode 100644 index 0000000000000000000000000000000000000000..fba1f4c122236011eb8429643b9cb43eeeebd85a --- /dev/null +++ b/physiological_integration/cases/case38/case38_21.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[40, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_22.json b/physiological_integration/cases/case38/case38_22.json new file mode 100644 index 0000000000000000000000000000000000000000..1916efa7b7474ce28c55e9c6314fde43f9fc53e6 --- /dev/null +++ b/physiological_integration/cases/case38/case38_22.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_23.json b/physiological_integration/cases/case38/case38_23.json new file mode 100644 index 0000000000000000000000000000000000000000..ce6060ce124e1c63bb8a866848d051c328a9f5bb --- /dev/null +++ b/physiological_integration/cases/case38/case38_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[10, 50)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_24.json b/physiological_integration/cases/case38/case38_24.json new file mode 100644 index 0000000000000000000000000000000000000000..14a9eb8063f7f5595bde88ab84eea36f91e63ef3 --- /dev/null +++ b/physiological_integration/cases/case38/case38_24.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "label_path": "data/multi/DEAP_03.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_25.json b/physiological_integration/cases/case38/case38_25.json new file mode 100644 index 0000000000000000000000000000000000000000..2e50c5a1e97c57b967b9d89b59d0ea0395d6d557 --- /dev/null +++ b/physiological_integration/cases/case38/case38_25.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_26.json b/physiological_integration/cases/case38/case38_26.json new file mode 100644 index 0000000000000000000000000000000000000000..586e7c0927a9fd7cb76d59640c234a8fb7c3e795 --- /dev/null +++ b/physiological_integration/cases/case38/case38_26.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 20)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_27.json b/physiological_integration/cases/case38/case38_27.json new file mode 100644 index 0000000000000000000000000000000000000000..e2f09aa884ee52401ddd433d7918b06d7e1a1f07 --- /dev/null +++ b/physiological_integration/cases/case38/case38_27.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[20, 40)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_28.json b/physiological_integration/cases/case38/case38_28.json new file mode 100644 index 0000000000000000000000000000000000000000..55718bf3b4aa82729205939b7fb292268b248a10 --- /dev/null +++ b/physiological_integration/cases/case38/case38_28.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[40, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_29.json b/physiological_integration/cases/case38/case38_29.json new file mode 100644 index 0000000000000000000000000000000000000000..f0b54d8cd5d3211491313bb51749a81c9072aa49 --- /dev/null +++ b/physiological_integration/cases/case38/case38_29.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_30.json b/physiological_integration/cases/case38/case38_30.json new file mode 100644 index 0000000000000000000000000000000000000000..3f6feeec46c1e7fcc2b52f2f76c3f2edd32d9169 --- /dev/null +++ b/physiological_integration/cases/case38/case38_30.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 45)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_31.json b/physiological_integration/cases/case38/case38_31.json new file mode 100644 index 0000000000000000000000000000000000000000..193a7909125489c6bcbdc4668129458713c4fa91 --- /dev/null +++ b/physiological_integration/cases/case38/case38_31.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "dominance" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_32.json b/physiological_integration/cases/case38/case38_32.json new file mode 100644 index 0000000000000000000000000000000000000000..5560f33d60e5aa2abcb298752475f5a4a682d3f1 --- /dev/null +++ b/physiological_integration/cases/case38/case38_32.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "label_path": "data/multi/DEAP_04.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_33.json b/physiological_integration/cases/case38/case38_33.json new file mode 100644 index 0000000000000000000000000000000000000000..39581e419a4355c938733554f6014fc91c1d1022 --- /dev/null +++ b/physiological_integration/cases/case38/case38_33.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 15)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_34.json b/physiological_integration/cases/case38/case38_34.json new file mode 100644 index 0000000000000000000000000000000000000000..e2acd71f9ea8ca700dc7b8fd29ad556a9a3cbda5 --- /dev/null +++ b/physiological_integration/cases/case38/case38_34.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 45)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_35.json b/physiological_integration/cases/case38/case38_35.json new file mode 100644 index 0000000000000000000000000000000000000000..c8c686b82a212dcc0b0bffcfb09d8804bf0f0f16 --- /dev/null +++ b/physiological_integration/cases/case38/case38_35.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[45, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "dominance" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_36.json b/physiological_integration/cases/case38/case38_36.json new file mode 100644 index 0000000000000000000000000000000000000000..ee37fae1547f5e50e42ce0baceb245b624d3a6ac --- /dev/null +++ b/physiological_integration/cases/case38/case38_36.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 20)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_37.json b/physiological_integration/cases/case38/case38_37.json new file mode 100644 index 0000000000000000000000000000000000000000..274f3621ce0eff73d622a11c2b2dce03ccc00af1 --- /dev/null +++ b/physiological_integration/cases/case38/case38_37.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[0, 30)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_38.json b/physiological_integration/cases/case38/case38_38.json new file mode 100644 index 0000000000000000000000000000000000000000..a5a49acb930906fbf5fa77a23a1b0fd09878a4ca --- /dev/null +++ b/physiological_integration/cases/case38/case38_38.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[15, 45)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_39.json b/physiological_integration/cases/case38/case38_39.json new file mode 100644 index 0000000000000000000000000000000000000000..007c7bb0a19b55906d33be8b41d064523dc086e2 --- /dev/null +++ b/physiological_integration/cases/case38/case38_39.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[30, 60)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "arousal" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case38/case38_40.json b/physiological_integration/cases/case38/case38_40.json new file mode 100644 index 0000000000000000000000000000000000000000..c5f7f97f00b1667569a4f908e0ab061115c84459 --- /dev/null +++ b/physiological_integration/cases/case38/case38_40.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case38_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 2, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "label_path": "data/multi/DEAP_05.npy", + "instruction": "Use all complete DEAP trials and the public `valence`, `arousal`, `dominance`, and `liking` ratings.\n\nFor each trial, use the 3 seconds before stimulus onset as baseline and `[10, 50)` seconds after stimulus onset as the response period.\n\nFor EEG, estimate alpha power in `[8, 13)` Hz from the homologous pairs `Fp1/Fp2`, `AF3/AF4`, and `F3/F4`. For each interval, compute right-minus-left log-alpha power for each pair and average across the three pairs. Define the EEG response as response-period asymmetry minus baseline asymmetry.\n\nFor `GSR1`, define the response as the response-period mean minus the baseline mean.\n\nCompute the Spearman correlation between each physiological response and each of the four ratings. For each modality, report the rating with the largest absolute correlation as `eeg_best_rating` and `gsr_best_rating`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON extraction engine for DEAP physiological-rating analyses.\n\n### TASK\nExtract:\n- eeg_best_rating: the rating with the largest absolute Spearman correlation with the trial-level EEG response.\n- gsr_best_rating: the rating with the largest absolute Spearman correlation with the trial-level GSR response.\n\n### STRICT CONSTRAINTS\n1. Output ONLY one valid JSON object.\n2. Do not output Markdown, explanations, correlation values, units, or extra fields.\n3. The keys must be exactly `eeg_best_rating` and `gsr_best_rating`.\n4. Each value must be exactly `valence`, `arousal`, `dominance`, `liking`, or null if missing or unclear.\n5. Extract the rating selected by absolute correlation magnitude, regardless of correlation sign.\n\n### OUTPUT TEMPLATE\n{\"eeg_best_rating\": \"\", \"gsr_best_rating\": \"\"}", + "metrics": [ + { + "metric_id": "eeg_best_rating", + "type": "categorical_check", + "target_key": "eeg_best_rating", + "weight": 50, + "params": { + "gt_value": "liking" + } + }, + { + "metric_id": "gsr_best_rating", + "type": "categorical_check", + "target_key": "gsr_best_rating", + "weight": 50, + "params": { + "gt_value": "valence" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_01.json b/physiological_integration/cases/case39/case39_01.json new file mode 100644 index 0000000000000000000000000000000000000000..15bfe54dcf98b1f1c8c631c15dcbee37850d69af --- /dev/null +++ b/physiological_integration/cases/case39/case39_01.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Compared with fMRI, which property of EEG directly supports resolving rapid neural events? Options: A: EEG inherently provides millimeter-scale cortical source localization. B: EEG provides millisecond-scale temporal resolution. C: EEG is more sensitive to deep nuclei than to cortical activity. D: EEG is unaffected by head models and volume conduction. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_02.json b/physiological_integration/cases/case39/case39_02.json new file mode 100644 index 0000000000000000000000000000000000000000..40dd2413684ab4c8d6f8603f664fa21d14d8d902 --- /dev/null +++ b/physiological_integration/cases/case39/case39_02.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which property limits exact source localization from scalp EEG potentials? Options: A: Frequency resolution removes spatial ambiguity without a head model. B: The sampling frequency alone determines every cortical source location. C: Volume conduction and the non-unique inverse problem limit exact source localization. D: Changing the reference produces a unique anatomical inverse solution. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_03.json b/physiological_integration/cases/case39/case39_03.json new file mode 100644 index 0000000000000000000000000000000000000000..8c4cfcd6f6ba3d4dfb19702957f132542b1549d0 --- /dev/null +++ b/physiological_integration/cases/case39/case39_03.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What electrical quantity does a conventional EEG channel record? Options: A: It records an absolute, reference-free electric potential. B: It records an optical change in cerebral blood volume. C: It records a voltage difference relative to a reference electrode or reference scheme. D: It counts action potentials from one identified neuron. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_04.json b/physiological_integration/cases/case39/case39_04.json new file mode 100644 index 0000000000000000000000000000000000000000..1d2118b93c067aedda712ecc4d8b405fd6a7041c --- /dev/null +++ b/physiological_integration/cases/case39/case39_04.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: A signal contains frequencies up to f_max. Which sampling condition supports alias-free representation of that band? Options: A: The sampling rate is strictly greater than 2 f_max. B: The sampling rate is lower than f_max. C: The sampling rate is determined only by recording duration. D: The sampling rate equals f_max. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_05.json b/physiological_integration/cases/case39/case39_05.json new file mode 100644 index 0000000000000000000000000000000000000000..11b2af9eb67d09bd16bdfeba1e25228d133fc309 --- /dev/null +++ b/physiological_integration/cases/case39/case39_05.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which operation is required before reducing the sampling rate of a band-limited physiological signal? Options: A: Re-referencing replaces the need for anti-alias filtering. B: The samples are discarded first and aliasing is removed afterward. C: A high-pass filter is applied after downsampling. D: A low-pass anti-aliasing filter is applied before downsampling. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_06.json b/physiological_integration/cases/case39/case39_06.json new file mode 100644 index 0000000000000000000000000000000000000000..8a13b5d42a49b4c72860b16d3b761b708cc8cbb8 --- /dev/null +++ b/physiological_integration/cases/case39/case39_06.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which procedure enhances a reproducible event-related potential across repeated trials? Options: A: Electrode coordinates are averaged instead of signal samples. B: Trials are averaged without aligning their event times. C: Only the largest single-trial peak is retained. D: Trials are aligned to the event and averaged across repetitions. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_07.json b/physiological_integration/cases/case39/case39_07.json new file mode 100644 index 0000000000000000000000000000000000000000..b6c3c6259156ca56f4d6bd5809b0c38ae0ff29d8 --- /dev/null +++ b/physiological_integration/cases/case39/case39_07.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What is the purpose of pre-event baseline correction in an ERP analysis? Options: A: It deletes all signal samples after the event. B: It expresses post-event activity relative to a pre-event reference level. C: It guarantees removal of every physiological artifact. D: It changes the recording sampling rate. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_08.json b/physiological_integration/cases/case39/case39_08.json new file mode 100644 index 0000000000000000000000000000000000000000..b06cb763db611d9d0652b398b6ddd0d2c521bbc5 --- /dev/null +++ b/physiological_integration/cases/case39/case39_08.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What does the power spectral density of an EEG segment describe? Options: A: It describes how signal power is distributed over frequency. B: It lists the spatial coordinates of the electrodes. C: It orders raw samples by their voltage magnitude. D: It lists the onset times of experimental events. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_09.json b/physiological_integration/cases/case39/case39_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e18bf9e4a9ddc14b0eadf132b03a23523f26998a --- /dev/null +++ b/physiological_integration/cases/case39/case39_09.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which physiological electrical source underlies electrooculography? Options: A: EOG uses an optical image of retinal blood vessels. B: EOG measures mechanical displacement of extraocular muscles. C: EOG directly measures pupil diameter with a camera. D: EOG uses the standing potential between the cornea and retina. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_10.json b/physiological_integration/cases/case39/case39_10.json new file mode 100644 index 0000000000000000000000000000000000000000..930864564491cfe239c291ca10f92b47db94ad71 --- /dev/null +++ b/physiological_integration/cases/case39/case39_10.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which channel configuration directly represents left-right eye movement? Options: A: A vertical EOG derivation directly represents only left-right movement. B: A chin EMG derivation directly measures eye position. C: An occipital EEG derivation directly measures gaze coordinates. D: A horizontal EOG derivation directly represents left-right eye movement. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_11.json b/physiological_integration/cases/case39/case39_11.json new file mode 100644 index 0000000000000000000000000000000000000000..50ca5eb080aa0df8071d05df8a70fd7625fb46d0 --- /dev/null +++ b/physiological_integration/cases/case39/case39_11.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which paired signal pattern is compatible with an eye blink? Options: A: A transient appears only in horizontal EOG and never in frontal EEG. B: A blink creates only an R peak in ECG. C: A transient appears in vertical EOG and frontal EEG channels. D: A blink creates only an occipital alpha oscillation. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_12.json b/physiological_integration/cases/case39/case39_12.json new file mode 100644 index 0000000000000000000000000000000000000000..27c0ca5b59a3617acfc8d4c4f1ccf48d583b9e24 --- /dev/null +++ b/physiological_integration/cases/case39/case39_12.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which statement distinguishes EOG from camera-based EyeTracking when the eyelids are closed? Options: A: EOG can retain an electrical eye-position signal while camera tracking loses visual access to the eye. B: Both methods require a visible pupil in every sample. C: EOG measures only pupil diameter through an optical lens. D: Camera tracking records the corneo-retinal electrical potential through closed eyelids. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_13.json b/physiological_integration/cases/case39/case39_13.json new file mode 100644 index 0000000000000000000000000000000000000000..3801860b0f5fc00ed5273b00ae2572db29bd5002 --- /dev/null +++ b/physiological_integration/cases/case39/case39_13.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How do gaze position and pupil diameter differ in EyeTracking data? Options: A: Both variables are alternative names for pupil size. B: Both variables are electrical EOG voltages. C: Gaze coordinates describe viewing location, while pupil diameter describes pupil size. D: Gaze coordinates describe pupil size, while pupil diameter describes screen location. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_14.json b/physiological_integration/cases/case39/case39_14.json new file mode 100644 index 0000000000000000000000000000000000000000..e87bc3a182485a77c7e5731459ac3837f6e81f08 --- /dev/null +++ b/physiological_integration/cases/case39/case39_14.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What does a validity field attached to a gaze or pupil sample indicate? Options: A: It encodes the participant's emotion category. B: It indicates whether the current EyeTracking sample is considered reliable. C: It reports the quality of an EEG reference electrode. D: It assigns a permanent identifier to a fixation. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_15.json b/physiological_integration/cases/case39/case39_15.json new file mode 100644 index 0000000000000000000000000000000000000000..68d168c90fd7334fa3d7d52f523df08b9cbbb1ff --- /dev/null +++ b/physiological_integration/cases/case39/case39_15.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What does surface EMG record at the skin above a muscle? Options: A: It directly records absolute mechanical force. B: It records sweat-gland conductance. C: It directly records muscle temperature. D: It records electrical activity generated by active muscle fibers and motor units. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_16.json b/physiological_integration/cases/case39/case39_16.json new file mode 100644 index 0000000000000000000000000000000000000000..920a9155a0ea3b71dd74ebaa44d20b7342902923 --- /dev/null +++ b/physiological_integration/cases/case39/case39_16.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Why is uncalibrated surface EMG amplitude not an absolute cross-subject measure of muscle force? Options: A: Crosstalk alters ECG but cannot alter surface EMG. B: Electrode geometry, tissue properties, and crosstalk affect the recorded amplitude. C: Recorded voltage is numerically identical to force in newtons. D: Electrode placement has no effect on surface EMG amplitude. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_17.json b/physiological_integration/cases/case39/case39_17.json new file mode 100644 index 0000000000000000000000000000000000000000..e27ec9f9cdaf21ee1338401fbdabeaef63577005 --- /dev/null +++ b/physiological_integration/cases/case39/case39_17.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which physiological mechanism directly drives electrodermal conductance changes? Options: A: Arterial oxygen saturation is the electrical source of EDA. B: Cortical voltage is measured directly by EDA electrodes. C: Sympathetically mediated sweat-gland activity changes skin conductance. D: Ventricular depolarization changes skin conductance directly. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_18.json b/physiological_integration/cases/case39/case39_18.json new file mode 100644 index 0000000000000000000000000000000000000000..6112a57076dc72051c65b6536d89e2cadd4ae57f --- /dev/null +++ b/physiological_integration/cases/case39/case39_18.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which distinction between tonic skin conductance and a phasic SCR is correct? Options: A: Tonic conductance is a slowly varying level, while a phasic SCR is a transient response. B: Both measures are intervals between PPG pulses. C: Both measures are slopes of peripheral temperature. D: Tonic conductance is transient, while a phasic SCR is the slow baseline. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_19.json b/physiological_integration/cases/case39/case39_19.json new file mode 100644 index 0000000000000000000000000000000000000000..0e9666b3e5b6ffaa2d9f738c9a17bd023fefef90 --- /dev/null +++ b/physiological_integration/cases/case39/case39_19.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What physiological quantity is detected by photoplethysmography? Options: A: It mechanically measures airflow through the airway. B: It electrically measures skeletal muscle motor units. C: It electrically detects myocardial depolarization. D: It optically detects changes in blood volume within the peripheral microvascular bed. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_20.json b/physiological_integration/cases/case39/case39_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0d9bf8d558359394e104d234752da41451ec7390 --- /dev/null +++ b/physiological_integration/cases/case39/case39_20.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which PPG measurement supports estimation of pulse rate? Options: A: The period of an EEG alpha oscillation is the PPG pulse interval. B: The distance between the two pupils determines pulse rate. C: The time interval between successive pulse peaks supports pulse-rate estimation. D: The slope of peripheral temperature directly defines pulse rate. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_21.json b/physiological_integration/cases/case39/case39_21.json new file mode 100644 index 0000000000000000000000000000000000000000..08b913d4fb415a0e3d733160f2e6ac2d79cf071f --- /dev/null +++ b/physiological_integration/cases/case39/case39_21.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Why is peripheral temperature unsuitable for localizing a millisecond-scale event onset? Options: A: Its sensor reports neuronal spikes at millisecond resolution. B: Its physiological response evolves too slowly to resolve millisecond event timing. C: It is the same electrical measurement as ECG. D: It directly contains an event-related cortical potential. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_22.json b/physiological_integration/cases/case39/case39_22.json new file mode 100644 index 0000000000000000000000000000000000000000..bac69a5653aef2027ad1dfa459ecd69b86e364f0 --- /dev/null +++ b/physiological_integration/cases/case39/case39_22.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What conclusion is justified from EDA, PPG, or peripheral temperature alone? Options: A: These signals characterize autonomic physiology but do not uniquely determine valence. B: A GSR increase uniquely identifies positive valence. C: The sign of a temperature slope uniquely identifies valence. D: A PPG amplitude change uniquely identifies negative valence. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_23.json b/physiological_integration/cases/case39/case39_23.json new file mode 100644 index 0000000000000000000000000000000000000000..1a573bf71869c80b4c601e9e2e2743b7acea2cf8 --- /dev/null +++ b/physiological_integration/cases/case39/case39_23.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which statement correctly distinguishes ECG from PPG? Options: A: ECG measures an optical pulse, while PPG measures cardiac electrical activity. B: ECG measures cardiac electrical activity, while PPG optically measures a peripheral blood-volume pulse. C: Both ECG and PPG are electrical measurements of myocardial voltage. D: Both ECG and PPG are optical measurements of blood volume. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_24.json b/physiological_integration/cases/case39/case39_24.json new file mode 100644 index 0000000000000000000000000000000000000000..2cdbdfd65ebedecb8980ee04ab1d82fc0037b2a4 --- /dev/null +++ b/physiological_integration/cases/case39/case39_24.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What does an ECG R-R interval represent? Options: A: It is the amplitude of one peripheral pulse. B: It is the voltage difference between a P wave and a Q wave. C: It is the duration of one respiratory cycle measured without ECG. D: It is the elapsed time between two successive R peaks. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_25.json b/physiological_integration/cases/case39/case39_25.json new file mode 100644 index 0000000000000000000000000000000000000000..f9d2d9a7d9949f491cf2025f765c7f65be46ed92 --- /dev/null +++ b/physiological_integration/cases/case39/case39_25.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What quantity does SpO2 represent? Options: A: It is a measurement of thoracic respiratory effort. B: It is a pulse-oximeter estimate of peripheral arterial oxygen saturation. C: It is a direct measurement of arterial blood pressure. D: It is a measurement of exhaled carbon-dioxide concentration. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_26.json b/physiological_integration/cases/case39/case39_26.json new file mode 100644 index 0000000000000000000000000000000000000000..f0b4623eaede1dbabe417b8ed1216e98b097b85a --- /dev/null +++ b/physiological_integration/cases/case39/case39_26.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_26", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which physical principle supports pulse-oximeter estimation of SpO2? Options: A: Thoracic displacement directly measures hemoglobin absorption. B: A single EMG electrode directly separates oxyhemoglobin from deoxyhemoglobin. C: Oxyhemoglobin and deoxyhemoglobin absorb selected light wavelengths differently. D: Scalp EEG voltage changes directly encode arterial saturation. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_27.json b/physiological_integration/cases/case39/case39_27.json new file mode 100644 index 0000000000000000000000000000000000000000..3148b16713a6b51864c09dd7e8ba5d03d8f32d11 --- /dev/null +++ b/physiological_integration/cases/case39/case39_27.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_27", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How do an airflow channel and thoracic or abdominal effort channels differ? Options: A: Airflow measures movement of air, while effort channels measure respiratory movement or effort. B: Effort channels measure arterial oxygen saturation. C: Airflow channels measure cortical electrical activity. D: The two channel types measure the same physical quantity. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_28.json b/physiological_integration/cases/case39/case39_28.json new file mode 100644 index 0000000000000000000000000000000000000000..b7ae59998af63f1b7ffeb216c6565e84f86d44f1 --- /dev/null +++ b/physiological_integration/cases/case39/case39_28.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_28", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which mapping of PSG signals to physiological evidence is correct? Options: A: All three signals measure arterial oxygen saturation. B: EEG provides eye movements, EOG provides chin tone, and chin EMG provides cortical activity. C: All three signals measure respiratory airflow. D: EEG provides brain electrical activity, EOG provides eye movements, and chin EMG provides chin muscle tone. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_29.json b/physiological_integration/cases/case39/case39_29.json new file mode 100644 index 0000000000000000000000000000000000000000..6dd0963483027a64d6ce47ede0fd771370c8fd1e --- /dev/null +++ b/physiological_integration/cases/case39/case39_29.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_29", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which physiological quantities are estimated by conventional functional near-infrared spectroscopy? Options: A: It directly records cardiac electrical depolarization. B: It directly measures absolute cerebral blood flow in every vessel. C: It estimates changes in oxyhemoglobin and deoxyhemoglobin concentration. D: It directly records neuronal action-potential voltage. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_30.json b/physiological_integration/cases/case39/case39_30.json new file mode 100644 index 0000000000000000000000000000000000000000..8ca8e724e9757fb17d2f8942f6e2071f9fb78726 --- /dev/null +++ b/physiological_integration/cases/case39/case39_30.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_30", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What is the role of the Modified Beer-Lambert Law in fNIRS processing? Options: A: It relates optical-density changes to changes in chromophore concentration. B: It determines the digitizer sampling rate. C: It converts EEG voltage directly into hemoglobin concentration. D: It guarantees complete removal of superficial physiology. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_31.json b/physiological_integration/cases/case39/case39_31.json new file mode 100644 index 0000000000000000000000000000000000000000..0f13f271a60ca59d75d4f09143cfb405d09b1b42 --- /dev/null +++ b/physiological_integration/cases/case39/case39_31.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_31", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How do task-related EEG and fNIRS response times differ physiologically? Options: A: Every observed EEG-fNIRS delay proves a device clock error. B: The fNIRS hemodynamic response follows the rapid EEG electrical response with a delay of seconds. C: The two physiological responses have an obligatory zero delay. D: The EEG electrical response is delayed by minutes relative to fNIRS. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_32.json b/physiological_integration/cases/case39/case39_32.json new file mode 100644 index 0000000000000000000000000000000000000000..11630c790c0f0d57db74e3e2e5b85b15ebd98838 --- /dev/null +++ b/physiological_integration/cases/case39/case39_32.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_32", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which tissue is within the characteristic sensitivity range of conventional scalp fNIRS? Options: A: Superficial cortical tissue beneath the source-detector array is within its sensitivity range. B: Only the skull contributes to every fNIRS channel. C: All brain depths contribute with identical sensitivity. D: Deep brainstem nuclei are sampled preferentially over cortex. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_33.json b/physiological_integration/cases/case39/case39_33.json new file mode 100644 index 0000000000000000000000000000000000000000..0a4e5e7409a7d36085791c1bd445218aad2cb81b --- /dev/null +++ b/physiological_integration/cases/case39/case39_33.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_33", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What measurement property is affected by fNIRS source-detector separation? Options: A: It changes only the acquisition timestamp format. B: It affects photon sampling depth and the spatial sensitivity profile. C: It has no relation to photon paths through tissue. D: It affects only the exported file name. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_34.json b/physiological_integration/cases/case39/case39_34.json new file mode 100644 index 0000000000000000000000000000000000000000..3ae992c0e433cc9585a7e88fbb9dee9d71cc9056 --- /dev/null +++ b/physiological_integration/cases/case39/case39_34.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_34", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What is the methodological role of a short-separation fNIRS channel? Options: A: It serves as the electrical reference for EEG. B: It selectively measures deep cortical activation. C: It stores the experimental condition label. D: It estimates superficial scalp and systemic physiological components for nuisance control. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_35.json b/physiological_integration/cases/case39/case39_35.json new file mode 100644 index 0000000000000000000000000000000000000000..e897fc887e6410245eaa2916ab984a13d8ab429a --- /dev/null +++ b/physiological_integration/cases/case39/case39_35.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_35", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: Which statement distinguishes neurovascular delay from acquisition clock offset? Options: A: Clock synchronization removes the biological hemodynamic response time. B: Physiological delay and clock offset are the same quantity. C: A physiological hemodynamic delay can remain after the device clocks and event times are correctly aligned. D: Every cross-modal lag is an acquisition clock error. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_36.json b/physiological_integration/cases/case39/case39_36.json new file mode 100644 index 0000000000000000000000000000000000000000..aaff9673af9d0d41eff051e0dedf9fb18cebbf89 --- /dev/null +++ b/physiological_integration/cases/case39/case39_36.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_36", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How should EEG and an auxiliary stream with a different sampling rate be aligned? Options: A: Samples are related through timestamps or a shared physical-time axis. B: Samples with the same integer index are treated as simultaneous. C: Both files are truncated to the same number of samples without time mapping. D: File-system ordering defines physiological simultaneity. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_37.json b/physiological_integration/cases/case39/case39_37.json new file mode 100644 index 0000000000000000000000000000000000000000..ec348477a1de9cf5869106824b84f5c297e9a02b --- /dev/null +++ b/physiological_integration/cases/case39/case39_37.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_37", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What causal conclusion follows from a high correlation between two physiological modalities alone? Options: A: High correlation proves that the first modality causes the second. B: Zero correlation proves that no relation of any kind exists. C: Correlation alone does not establish that either modality causes the other. D: Correlation uniquely identifies the biological mechanism linking the modalities. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_38.json b/physiological_integration/cases/case39/case39_38.json new file mode 100644 index 0000000000000000000000000000000000000000..10d5c0467f26afdcccfe720ef557b0b996017ce4 --- /dev/null +++ b/physiological_integration/cases/case39/case39_38.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_38", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: What capability is tested by leave-one-subject-out evaluation? Options: A: It evaluates the model on the same observations used for fitting. B: It tests interpolation among samples from the same training subject. C: It resamples time points while retaining every subject in training. D: It tests generalization to a subject excluded from model fitting and selection. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_39.json b/physiological_integration/cases/case39/case39_39.json new file mode 100644 index 0000000000000000000000000000000000000000..ba6146e79395a2cb46ad710c0729e6434b49b520 --- /dev/null +++ b/physiological_integration/cases/case39/case39_39.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_39", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How should an unobserved physiological sample be represented when zero is a valid measured value? Options: A: Missingness is preserved with a validity mask or an explicit missing-value representation. B: Every missing sample is replaced by zero without an indicator. C: The time axis is deleted wherever one modality has a missing sample. D: A neighboring sample is copied and the replacement is left unmarked. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} diff --git a/physiological_integration/cases/case39/case39_40.json b/physiological_integration/cases/case39/case39_40.json new file mode 100644 index 0000000000000000000000000000000000000000..f3429abb1d8583c9ac593eaa5c45dfc1eb93f67b --- /dev/null +++ b/physiological_integration/cases/case39/case39_40.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "case39_40", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "Multi_QA" + }, + "agent_input": { + "data_path": [], + "instruction": "Answer this standalone multimodal neurophysiology multiple-choice question: How should a fusion system treat one modality whose current signal quality is poor? Options: A: A mechanical majority vote is used without checking modality reliability. B: A quality-aware rule reduces the contribution of evidence from the unreliable modality. C: Every modality retains an identical fixed weight regardless of quality. D: The unreliable modality receives a larger weight because it is noisy. Select exactly one option, A, B, C, or D. Return only one JSON object with the single field answer and no explanation or extra fields." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the answer choice from the agent's report.\n- answer: exactly one of A, B, C, or D.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON key must be EXACTLY \"answer\".\n5. The answer value must be exactly one of \"A\", \"B\", \"C\", or \"D\".\n6. If the answer is missing or unclear, return {\"answer\": null}.\n\n### OUTPUT TEMPLATE\n{\"answer\": \"\"}", + "metrics": [ + { + "metric_id": "knowledge_answer_accuracy", + "type": "categorical_check", + "target_key": "answer", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_01.json b/physiological_integration/cases/case4/case4_01.json new file mode 100644 index 0000000000000000000000000000000000000000..824052822cb432980e2ec5d7f91ea90bd95b18b8 --- /dev/null +++ b/physiological_integration/cases/case4/case4_01.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Report the native sampling rates of `Fp1` as `signal_1_sampling_rate` and `GSR1` as `signal_2_sampling_rate`. Then compute the mean `GSR1` value in `µS` over the first second of the recording, using the `GSR1` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 5.280209, + "tolerance": 0.26401045 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_02.json b/physiological_integration/cases/case4/case4_02.json new file mode 100644 index 0000000000000000000000000000000000000000..6d4929b4f26ac81aaa569ed348cd866d13f1c855 --- /dev/null +++ b/physiological_integration/cases/case4/case4_02.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Report the native sampling rates of `Resp` as `signal_1_sampling_rate` and `Plet` as `signal_2_sampling_rate`. Then compute the mean `Plet` value in `mV` over the first second of the recording, using the `Plet` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -27.93517, + "tolerance": 1.3967585 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_03.json b/physiological_integration/cases/case4/case4_03.json new file mode 100644 index 0000000000000000000000000000000000000000..960d0f069979794bda2bed5313f4b57ef2a25451 --- /dev/null +++ b/physiological_integration/cases/case4/case4_03.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Report the native sampling rates of `Fz` as `signal_1_sampling_rate` and `Temp` as `signal_2_sampling_rate`. Then compute the mean `Fz` value in `uV` over the first second of the recording, using the `Fz` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 2031.855864, + "tolerance": 101.59279320000002 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_04.json b/physiological_integration/cases/case4/case4_04.json new file mode 100644 index 0000000000000000000000000000000000000000..e66c0e213f6ed8cd0a6c21bd3562f3f2ae9dd9ae --- /dev/null +++ b/physiological_integration/cases/case4/case4_04.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Report the native sampling rates of `GSR2` as `signal_1_sampling_rate` and `Resp` as `signal_2_sampling_rate`. Then compute the mean `GSR2` value in `nS` over the last second of the recording, using the `GSR2` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -0.484375, + "tolerance": 0.02421875 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_05.json b/physiological_integration/cases/case4/case4_05.json new file mode 100644 index 0000000000000000000000000000000000000000..041975303856098f300044e7dd12536b1c189665 --- /dev/null +++ b/physiological_integration/cases/case4/case4_05.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Report the native sampling rates of `Cz` as `signal_1_sampling_rate` and `Plet` as `signal_2_sampling_rate`. Then compute the mean `Plet` value in `V` over the last second of the recording, using the `Plet` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 512.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -0.028248, + "tolerance": 0.0014124 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_06.json b/physiological_integration/cases/case4/case4_06.json new file mode 100644 index 0000000000000000000000000000000000000000..03940299306417374e450af6133bcbe002b11934 --- /dev/null +++ b/physiological_integration/cases/case4/case4_06.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Report the native sampling rates of `FT7` as `signal_1_sampling_rate` and `eog_h` as `signal_2_sampling_rate`. Then compute the mean `FT7` value in `dataset_native_amplitude` over the first second of the recording, using the `FT7` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 200.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 16.763955, + "tolerance": 0.83819775 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_07.json b/physiological_integration/cases/case4/case4_07.json new file mode 100644 index 0000000000000000000000000000000000000000..ef3a50f9813f5e53d786064431436e7cc7e70ca8 --- /dev/null +++ b/physiological_integration/cases/case4/case4_07.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Report the native sampling rates of `OZ` as `signal_1_sampling_rate` and `eog_v` as `signal_2_sampling_rate`. Then compute the mean `eog_v` value in `milli_dataset_native_amplitude` over the first second of the recording, using the `eog_v` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 200.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 24140.59639, + "tolerance": 1207.0298195 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_08.json b/physiological_integration/cases/case4/case4_08.json new file mode 100644 index 0000000000000000000000000000000000000000..087e7fec27f08d743a66a2e6e911db956f696da8 --- /dev/null +++ b/physiological_integration/cases/case4/case4_08.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Report the native sampling rates of `POZ` as `signal_1_sampling_rate` and `eog_h` as `signal_2_sampling_rate`. Then compute the mean `POZ` value in `kilo_dataset_native_amplitude` over the last second of the recording, using the `POZ` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 200.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 0.015064, + "tolerance": 0.0007532 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_09.json b/physiological_integration/cases/case4/case4_09.json new file mode 100644 index 0000000000000000000000000000000000000000..5c00ac1a47609adb89a6963c4212e4f10b8d62e9 --- /dev/null +++ b/physiological_integration/cases/case4/case4_09.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Report the native sampling rates of `O2` as `signal_1_sampling_rate` and `eog_v` as `signal_2_sampling_rate`. Then compute the mean `O2` value in `micro_dataset_native_amplitude` over the first second of the recording, using the `O2` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 200.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 2731829.881668, + "tolerance": 136591.4940834 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_10.json b/physiological_integration/cases/case4/case4_10.json new file mode 100644 index 0000000000000000000000000000000000000000..c4841739c6bb2a9fda5481c11b14adf9452c272f --- /dev/null +++ b/physiological_integration/cases/case4/case4_10.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Report the native sampling rates of `CP1` as `signal_1_sampling_rate` and `eog_h` as `signal_2_sampling_rate`. Then compute the mean `CP1` value in `centi_dataset_native_amplitude` over the last second of the recording, using the `CP1` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 200.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -448.584557, + "tolerance": 22.429227850000004 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_11.json b/physiological_integration/cases/case4/case4_11.json new file mode 100644 index 0000000000000000000000000000000000000000..8237498bbe55081a2e8704165cdb7fa3128c7e2a --- /dev/null +++ b/physiological_integration/cases/case4/case4_11.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Report the native sampling rates of `EEG(sec)` as `signal_1_sampling_rate` and `EOG(L)` as `signal_2_sampling_rate`. Then compute the mean `EEG(sec)` value in `mV` over the last second of the recording, using the `EEG(sec)` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 50.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 0.004686, + "tolerance": 0.0002343 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_12.json b/physiological_integration/cases/case4/case4_12.json new file mode 100644 index 0000000000000000000000000000000000000000..79ae75a3a77a67582370feee01abe67df44c81f5 --- /dev/null +++ b/physiological_integration/cases/case4/case4_12.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Report the native sampling rates of `THOR RES` as `signal_1_sampling_rate` and `SaO2` as `signal_2_sampling_rate`. Then compute the mean `THOR RES` value in `deci_dataset_native_amplitude` over the last second of the recording, using the `THOR RES` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 0.007843, + "tolerance": 0.00039214999999999996 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_13.json b/physiological_integration/cases/case4/case4_13.json new file mode 100644 index 0000000000000000000000000000000000000000..eaed5200dc1993613960018ad1647bceb54435aa --- /dev/null +++ b/physiological_integration/cases/case4/case4_13.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Report the native sampling rates of `ECG` as `signal_1_sampling_rate` and `H.R.` as `signal_2_sampling_rate`. Then compute the mean `ECG` value in `uV` over the last second of the recording, using the `ECG` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 6.078431, + "tolerance": 0.30392155000000004 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_14.json b/physiological_integration/cases/case4/case4_14.json new file mode 100644 index 0000000000000000000000000000000000000000..b9a431d8f8a5a0e7679466a573540fc424b3de08 --- /dev/null +++ b/physiological_integration/cases/case4/case4_14.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Report the native sampling rates of `EEG` as `signal_1_sampling_rate` and `ABDO RES` as `signal_2_sampling_rate`. Then compute the mean `ABDO RES` value in `hecto_dataset_native_amplitude` over the first second of the recording, using the `ABDO RES` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -0.003867, + "tolerance": 0.00019335 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_15.json b/physiological_integration/cases/case4/case4_15.json new file mode 100644 index 0000000000000000000000000000000000000000..9616461e1371f6f7e845274fbe442dc6ecc542d5 --- /dev/null +++ b/physiological_integration/cases/case4/case4_15.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "case04_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Report the native sampling rates of `EOG(R)` as `signal_1_sampling_rate` and `EMG` as `signal_2_sampling_rate`. Then compute the mean `EOG(R)` value in `nV` over the last second of the recording, using the `EOG(R)` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 50.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 125.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -21333.333333, + "tolerance": 1066.66666665 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_16.json b/physiological_integration/cases/case4/case4_16.json new file mode 100644 index 0000000000000000000000000000000000000000..c6a3a07a9802c67eaaf3fffda101ac9eb4051455 --- /dev/null +++ b/physiological_integration/cases/case4/case4_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Report the native sampling rates of `FP1` as `signal_1_sampling_rate` and `S01D01_760` as `signal_2_sampling_rate`. Then compute the mean `FP1` value in `V` over the last second of the recording, using the `FP1` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.416667, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -0.00012, + "tolerance": 6e-06 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_17.json b/physiological_integration/cases/case4/case4_17.json new file mode 100644 index 0000000000000000000000000000000000000000..c7225bd34833d7d7c3e84f0e988bb3f3ebb5062d --- /dev/null +++ b/physiological_integration/cases/case4/case4_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Report the native sampling rates of `HEOG` as `signal_1_sampling_rate` and `S01D01_850` as `signal_2_sampling_rate`. Then compute the mean `S01D01_850` value in `mV` over the last second of the recording, using the `S01D01_850` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.416667, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 0.000288, + "tolerance": 1.4400000000000001e-05 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_18.json b/physiological_integration/cases/case4/case4_18.json new file mode 100644 index 0000000000000000000000000000000000000000..6a281a61c7c2c76f1e9fa49542b2bf738dfa6c61 --- /dev/null +++ b/physiological_integration/cases/case4/case4_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Report the native sampling rates of `O2` as `signal_1_sampling_rate` and `VEOG` as `signal_2_sampling_rate`. Then compute the mean `O2` value in `uV` over the last second of the recording, using the `O2` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 67.77713, + "tolerance": 3.3888565 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_19.json b/physiological_integration/cases/case4/case4_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5e8179732eccb280b31c2c5279011ad35f6d7212 --- /dev/null +++ b/physiological_integration/cases/case4/case4_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Report the native sampling rates of `S01D01_760` as `signal_1_sampling_rate` and `Cz` as `signal_2_sampling_rate`. Then compute the mean `Cz` value in `nV` over the first second of the recording, using the `Cz` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.416667, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 13412.725181, + "tolerance": 670.63625905 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_20.json b/physiological_integration/cases/case4/case4_20.json new file mode 100644 index 0000000000000000000000000000000000000000..93cdb709be9546384b850d9545b0cb31e6d7de02 --- /dev/null +++ b/physiological_integration/cases/case4/case4_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Report the native sampling rates of `S01D01_850` as `signal_1_sampling_rate` and `HEOG` as `signal_2_sampling_rate`. Then compute the mean `HEOG` value in `pV` over the first second of the recording, using the `HEOG` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 10.416667, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -52416878.048036, + "tolerance": 2620843.9024018003 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_21.json b/physiological_integration/cases/case4/case4_21.json new file mode 100644 index 0000000000000000000000000000000000000000..a305ab4818471162af95259cb262b20c2065ea43 --- /dev/null +++ b/physiological_integration/cases/case4/case4_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Report the native sampling rates of `Gaze point X` as `signal_1_sampling_rate` and `FP1` as `signal_2_sampling_rate`. Then compute the mean `FP1` value in `mV` over the first second of the recording, using the `FP1` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 250.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -0.031818, + "tolerance": 0.0015909000000000001 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_22.json b/physiological_integration/cases/case4/case4_22.json new file mode 100644 index 0000000000000000000000000000000000000000..067a47da2ff029d25b5a3be900ff09aab4b062e9 --- /dev/null +++ b/physiological_integration/cases/case4/case4_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Report the native sampling rates of `Pupil diameter left` as `signal_1_sampling_rate` and `HEO` as `signal_2_sampling_rate`. Then compute the mean `HEO` value in `uV` over the first second of the recording, using the `HEO` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 250.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 195.787013, + "tolerance": 9.789350650000001 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_23.json b/physiological_integration/cases/case4/case4_23.json new file mode 100644 index 0000000000000000000000000000000000000000..a36e08ad811d695925eb13bd95ca39ebf7d98636 --- /dev/null +++ b/physiological_integration/cases/case4/case4_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Report the native sampling rates of `Gaze point Y` as `signal_1_sampling_rate` and `ECG` as `signal_2_sampling_rate`. Then compute the mean `ECG` value in `nV` over the first second of the recording, using the `ECG` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 250.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 24899.661541, + "tolerance": 1244.9830770500002 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_24.json b/physiological_integration/cases/case4/case4_24.json new file mode 100644 index 0000000000000000000000000000000000000000..3c1844a9eefc4c899483a14dd9bd31bc4db6af8e --- /dev/null +++ b/physiological_integration/cases/case4/case4_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "instruction": "Report the native sampling rates of `CZ` as `signal_1_sampling_rate` and `Pupil diameter right` as `signal_2_sampling_rate`. Then compute the mean `CZ` value in `V` over the first second of the recording, using the `CZ` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 250.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": -4e-06, + "tolerance": 2e-07 + } + } + ] + } +} diff --git a/physiological_integration/cases/case4/case4_25.json b/physiological_integration/cases/case4/case4_25.json new file mode 100644 index 0000000000000000000000000000000000000000..09430f24fe9b34e0fdfa5a7b483a002fbbc97bc6 --- /dev/null +++ b/physiological_integration/cases/case4/case4_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "case04_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "instruction": "Report the native sampling rates of `Pupil diameter right` as `signal_1_sampling_rate` and `VEO` as `signal_2_sampling_rate`. Then compute the mean `VEO` value in `pV` over the first second of the recording, using the `VEO` signal's own sampling rate, and report it as `selected_boundary_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_sampling_rate: Signal 1 sampling rate.\n- signal_2_sampling_rate: Signal 2 sampling rate.\n- selected_boundary_value: Selected boundary value. The arithmetic mean (average) of the selected signal over the requested boundary second, in the declared unit.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_sampling_rate\": , \"signal_2_sampling_rate\": , \"selected_boundary_value\": }", + "metrics": [ + { + "metric_id": "signal_1_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_1_sampling_rate", + "weight": 30, + "params": { + "gt_value": 250.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "signal_2_sampling_rate_accuracy", + "type": "numeric_check", + "target_key": "signal_2_sampling_rate", + "weight": 30, + "params": { + "gt_value": 1000.0, + "tolerance": 0.0 + } + }, + { + "metric_id": "selected_boundary_value_accuracy", + "type": "numeric_check", + "target_key": "selected_boundary_value", + "weight": 40, + "params": { + "gt_value": 2775728.702545, + "tolerance": 138786.43512725 + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_01.json b/physiological_integration/cases/case5/case5_01.json new file mode 100644 index 0000000000000000000000000000000000000000..19cc02a4f09441d142ed305b104624b1009602aa --- /dev/null +++ b/physiological_integration/cases/case5/case5_01.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Extract Fp1 (EEG) and GSR1 (GSR/EDA) from the same declared real-time interval [0, 1) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_01_signal_1.edf and the second at file_check/case05_01_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/DEAP_01.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_01_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Fp1" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.006529011367690037, + "weight": 25, + "tolerance": 0.0006529011367690037 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_01_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "GSR1" + ], + "weight": 25 + }, + "signal_rms": { + "value": 3948.8661442879984, + "weight": 25, + "tolerance": 394.8866144287999 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_02.json b/physiological_integration/cases/case5/case5_02.json new file mode 100644 index 0000000000000000000000000000000000000000..502475db3aaadd95370f81de9b5e374928caaeb9 --- /dev/null +++ b/physiological_integration/cases/case5/case5_02.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Extract Resp (Respiration) and Plet (PPG/BVP) from the same declared real-time interval [5, 7) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_02_signal_1.edf and the second at file_check/case05_02_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/DEAP_02.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_02_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Resp" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.04285457990540631, + "weight": 25, + "tolerance": 0.004285457990540631 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_02_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Plet" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.01671554772251551, + "weight": 25, + "tolerance": 0.0016715547722515511 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_03.json b/physiological_integration/cases/case5/case5_03.json new file mode 100644 index 0000000000000000000000000000000000000000..6e3be0440c006092a16cde7421817a07e561f2d6 --- /dev/null +++ b/physiological_integration/cases/case5/case5_03.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Extract Fz (EEG) and Temp (Temperature) from the same declared real-time interval [10, 13) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_03_signal_1.edf and the second at file_check/case05_03_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/DEAP_03.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_03_signal_1.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Fz" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.0020310686550348224, + "weight": 25, + "tolerance": 0.00020310686550348224 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_03_signal_2.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Temp" + ], + "weight": 25 + }, + "signal_rms": { + "value": 35.19301524479682, + "weight": 25, + "tolerance": 3.519301524479682 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_04.json b/physiological_integration/cases/case5/case5_04.json new file mode 100644 index 0000000000000000000000000000000000000000..5ddba9a3777ee2732849a1dc57d3fb857d3ee6e9 --- /dev/null +++ b/physiological_integration/cases/case5/case5_04.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Extract GSR2 (GSR/EDA) and Resp (Respiration) from the same declared real-time interval [15, 16) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_04_signal_1.edf and the second at file_check/case05_04_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/DEAP_04.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_04_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "GSR2" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.4843750288709998, + "weight": 25, + "tolerance": 0.048437502887099984 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_04_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Resp" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.043919183285745254, + "weight": 25, + "tolerance": 0.004391918328574526 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_05.json b/physiological_integration/cases/case5/case5_05.json new file mode 100644 index 0000000000000000000000000000000000000000..04e284d92171234fe412fa8a8a8b95f8d7df4340 --- /dev/null +++ b/physiological_integration/cases/case5/case5_05.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Extract Cz (EEG) and Plet (PPG/BVP) from the same declared real-time interval [20, 22) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_05_signal_1.edf and the second at file_check/case05_05_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/DEAP_05.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_05_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Cz" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.004376299611324381, + "weight": 25, + "tolerance": 0.0004376299611324381 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_05_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 512.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Plet" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.032086713195527666, + "weight": 25, + "tolerance": 0.003208671319552767 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_06.json b/physiological_integration/cases/case5/case5_06.json new file mode 100644 index 0000000000000000000000000000000000000000..cb91b33d25daad279d7e3a9e4a25e397cad1c238 --- /dev/null +++ b/physiological_integration/cases/case5/case5_06.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Extract FT7 (EEG) and eog_h (EOG) from the same declared real-time interval [0, 1) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_06_signal_1.edf and the second at file_check/case05_06_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_06_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 200.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "FT7" + ], + "weight": 25 + }, + "signal_rms": { + "value": 2.2466934979881513e-05, + "weight": 25, + "tolerance": 2.2466934979881512e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_06_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "eog_h" + ], + "weight": 25 + }, + "signal_rms": { + "value": 5.808972489619247e-05, + "weight": 25, + "tolerance": 5.808972489619247e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_07.json b/physiological_integration/cases/case5/case5_07.json new file mode 100644 index 0000000000000000000000000000000000000000..9c9551239c154374c9b7876fedf8834617a3fff4 --- /dev/null +++ b/physiological_integration/cases/case5/case5_07.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Extract OZ (EEG) and eog_v (EOG) from the same declared real-time interval [8, 10) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_07_signal_1.edf and the second at file_check/case05_07_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SEEDVIG_02_session-20151106_noon.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_07_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 200.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "OZ" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.8019072371053382e-05, + "weight": 25, + "tolerance": 1.8019072371053382e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_07_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "eog_v" + ], + "weight": 25 + }, + "signal_rms": { + "value": 8.972818558066559e-05, + "weight": 25, + "tolerance": 8.972818558066559e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_08.json b/physiological_integration/cases/case5/case5_08.json new file mode 100644 index 0000000000000000000000000000000000000000..9cfa123b78f3b5f7d677d9d305b207f1b7e8215a --- /dev/null +++ b/physiological_integration/cases/case5/case5_08.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Extract POZ (EEG) and eog_h (EOG) from the same declared real-time interval [16, 19) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_08_signal_1.edf and the second at file_check/case05_08_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SEEDVIG_03_session-20151024_noon.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_08_signal_1.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 200.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "POZ" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.1238381161134993e-05, + "weight": 25, + "tolerance": 1.1238381161134994e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_08_signal_2.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "eog_h" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.00023555626157439855, + "weight": 25, + "tolerance": 2.3555626157439857e-05 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_09.json b/physiological_integration/cases/case5/case5_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e6eb44eab5be3891e481f94ecaa9b38cca2fef25 --- /dev/null +++ b/physiological_integration/cases/case5/case5_09.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Extract O2 (EEG) and eog_v (EOG) from the same declared real-time interval [24, 25) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_09_signal_1.edf and the second at file_check/case05_09_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SEEDVIG_04_session-20151105_noon.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_09_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 200.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "O2" + ], + "weight": 25 + }, + "signal_rms": { + "value": 2.8974315038368e-05, + "weight": 25, + "tolerance": 2.8974315038368e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_09_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "eog_v" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.00018494590090313643, + "weight": 25, + "tolerance": 1.8494590090313645e-05 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_10.json b/physiological_integration/cases/case5/case5_10.json new file mode 100644 index 0000000000000000000000000000000000000000..9280fe89bc47b311c3df64cf381d4bc02eca91cf --- /dev/null +++ b/physiological_integration/cases/case5/case5_10.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Extract CP1 (EEG) and eog_h (EOG) from the same declared real-time interval [32, 34) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_10_signal_1.edf and the second at file_check/case05_10_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SEEDVIG_05_session-20141108_noon.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_10_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 200.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "CP1" + ], + "weight": 25 + }, + "signal_rms": { + "value": 8.47455643304692e-06, + "weight": 25, + "tolerance": 8.474556433046921e-07 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_10_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "eog_h" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.1015921806781855e-05, + "weight": 25, + "tolerance": 1.1015921806781856e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_11.json b/physiological_integration/cases/case5/case5_11.json new file mode 100644 index 0000000000000000000000000000000000000000..06e86abdb1360331057247aacd166be2dc8cca90 --- /dev/null +++ b/physiological_integration/cases/case5/case5_11.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Extract EEG(sec) (EEG) and EOG(L) (EOG) from the same declared real-time interval [304, 305) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_11_signal_1.edf and the second at file_check/case05_11_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SHHS1_01.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_11_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "EEG(sec)" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.6386632430565347e-05, + "weight": 25, + "tolerance": 1.6386632430565348e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_11_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 50.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "EOG(L)" + ], + "weight": 25 + }, + "signal_rms": { + "value": 4.430732645104557e-05, + "weight": 25, + "tolerance": 4.430732645104557e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_12.json b/physiological_integration/cases/case5/case5_12.json new file mode 100644 index 0000000000000000000000000000000000000000..4184e4556aacc4c9863e69abecedaafd6d791895 --- /dev/null +++ b/physiological_integration/cases/case5/case5_12.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Extract THOR RES (Respiration) and SaO2 (SaO2/SpO2) from the same declared real-time interval [13, 15) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_12_signal_1.edf and the second at file_check/case05_12_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SHHS1_02.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_12_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "THOR RES" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.06698884090536204, + "weight": 25, + "tolerance": 0.006698884090536205 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_12_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "SaO2" + ], + "weight": 25 + }, + "signal_rms": { + "value": 94.14053559166858, + "weight": 25, + "tolerance": 9.414053559166858 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_13.json b/physiological_integration/cases/case5/case5_13.json new file mode 100644 index 0000000000000000000000000000000000000000..217e322fa0e8f0024ff3a4b285dc53518a6b6437 --- /dev/null +++ b/physiological_integration/cases/case5/case5_13.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Extract ECG (ECG) and H.R. (HeartRate) from the same declared real-time interval [443, 446) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_13_signal_1.edf and the second at file_check/case05_13_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SHHS1_03.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_13_signal_1.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "ECG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.00024464520259057424, + "weight": 25, + "tolerance": 2.4464520259057425e-05 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_13_signal_2.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "H.R." + ], + "weight": 25 + }, + "signal_rms": { + "value": 87.24296498847212, + "weight": 25, + "tolerance": 8.724296498847213 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_14.json b/physiological_integration/cases/case5/case5_14.json new file mode 100644 index 0000000000000000000000000000000000000000..3f81ab379479dfdabd7a4194cccc1ef47ea01ec1 --- /dev/null +++ b/physiological_integration/cases/case5/case5_14.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Extract EEG (EEG) and ABDO RES (Respiration) from the same declared real-time interval [17, 18) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_14_signal_1.edf and the second at file_check/case05_14_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SHHS1_04.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_14_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "EEG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.1958935890601838e-05, + "weight": 25, + "tolerance": 1.1958935890601838e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_14_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "ABDO RES" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.7742511357580347, + "weight": 25, + "tolerance": 0.07742511357580348 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_15.json b/physiological_integration/cases/case5/case5_15.json new file mode 100644 index 0000000000000000000000000000000000000000..94b9de83219ef7edf8464ed498ec0dfe27e6957d --- /dev/null +++ b/physiological_integration/cases/case5/case5_15.json @@ -0,0 +1,88 @@ +{ + "meta_info": { + "case_id": "case05_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Extract EOG(R) (EOG) and EMG (EMG) from the same declared real-time interval [212, 214) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_15_signal_1.edf and the second at file_check/case05_15_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": "data/multi/SHHS1_05.npy" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_15_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 50.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "EOG(R)" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.5951495586844884e-05, + "weight": 25, + "tolerance": 1.5951495586844886e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_15_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 125.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "EMG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 2.0730753802269316e-05, + "weight": 25, + "tolerance": 2.0730753802269317e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_16.json b/physiological_integration/cases/case5/case5_16.json new file mode 100644 index 0000000000000000000000000000000000000000..c8e7fec60544bd55c05a4e3ada509f35a15ab039 --- /dev/null +++ b/physiological_integration/cases/case5/case5_16.json @@ -0,0 +1,94 @@ +{ + "meta_info": { + "case_id": "case05_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.edf", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.edf" + ], + "instruction": "Extract FP1 (EEG) and S01D01_760 (fNIRS) from the same declared real-time interval [46, 47) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_16_signal_1.edf and the second at file_check/case05_16_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/simultaneous_01_run-gonogo1_EEG.npy", + "data/multi/simultaneous_01_run-gonogo1_fNIRS.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_16_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "FP1" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.00012334421106858462, + "weight": 25, + "tolerance": 1.2334421106858464e-05 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_16_signal_2.edf" + }, + "duration_sec": { + "value": 0.9599999999999999, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.416666666666668, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "S01D01_760" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.17285220239368135, + "weight": 25, + "tolerance": 0.017285220239368137 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_17.json b/physiological_integration/cases/case5/case5_17.json new file mode 100644 index 0000000000000000000000000000000000000000..6d89ac1de6b9960f9d44b09d4152585cc7022506 --- /dev/null +++ b/physiological_integration/cases/case5/case5_17.json @@ -0,0 +1,94 @@ +{ + "meta_info": { + "case_id": "case05_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Extract HEOG (EOG) and S01D01_850 (fNIRS) from the same declared real-time interval [44, 46) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_17_signal_1.edf and the second at file_check/case05_17_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.npy", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_17_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "HEOG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 2.6183439061740344e-05, + "weight": 25, + "tolerance": 2.6183439061740346e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_17_signal_2.edf" + }, + "duration_sec": { + "value": 2.0159999999999996, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.416666666666668, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "S01D01_850" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.2831871316164416, + "weight": 25, + "tolerance": 0.028318713161644162 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_18.json b/physiological_integration/cases/case5/case5_18.json new file mode 100644 index 0000000000000000000000000000000000000000..bfb77ca5049dd3068385c31c63f28db356df1598 --- /dev/null +++ b/physiological_integration/cases/case5/case5_18.json @@ -0,0 +1,94 @@ +{ + "meta_info": { + "case_id": "case05_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.edf", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.edf" + ], + "instruction": "Extract O2 (EEG) and VEOG (EOG) from the same declared real-time interval [42, 45) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_18_signal_1.edf and the second at file_check/case05_18_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/simultaneous_03_run-gonogo1_EEG.npy", + "data/multi/simultaneous_03_run-gonogo1_fNIRS.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_18_signal_1.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "O2" + ], + "weight": 25 + }, + "signal_rms": { + "value": 7.738494882787959e-05, + "weight": 25, + "tolerance": 7.738494882787959e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_18_signal_2.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "VEOG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.00017282077933732077, + "weight": 25, + "tolerance": 1.7282077933732077e-05 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_19.json b/physiological_integration/cases/case5/case5_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ac0fc5d055b65a9b3c63d3b06e9ab512d79c0eae --- /dev/null +++ b/physiological_integration/cases/case5/case5_19.json @@ -0,0 +1,94 @@ +{ + "meta_info": { + "case_id": "case05_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Extract S01D01_760 (fNIRS) and Cz (EEG) from the same declared real-time interval [42, 43) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_19_signal_1.edf and the second at file_check/case05_19_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.npy", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_19_signal_1.edf" + }, + "duration_sec": { + "value": 0.9599999999999999, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.416666666666668, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "S01D01_760" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.1099768757343457, + "weight": 25, + "tolerance": 0.01099768757343457 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_19_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Cz" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.408960832022274e-05, + "weight": 25, + "tolerance": 1.4089608320222742e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_20.json b/physiological_integration/cases/case5/case5_20.json new file mode 100644 index 0000000000000000000000000000000000000000..716b74e2b7e41341c21bb6ec2cff61f7d33efedf --- /dev/null +++ b/physiological_integration/cases/case5/case5_20.json @@ -0,0 +1,94 @@ +{ + "meta_info": { + "case_id": "case05_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Extract S01D01_850 (fNIRS) and HEOG (EOG) from the same declared real-time interval [62, 64) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_20_signal_1.edf and the second at file_check/case05_20_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.npy", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_20_signal_1.edf" + }, + "duration_sec": { + "value": 1.9199999999999997, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 10.416666666666668, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "S01D01_850" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.6446572845162978, + "weight": 25, + "tolerance": 0.06446572845162978 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_20_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "HEOG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 9.741366536581087e-05, + "weight": 25, + "tolerance": 9.741366536581088e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_21.json b/physiological_integration/cases/case5/case5_21.json new file mode 100644 index 0000000000000000000000000000000000000000..24117c0c7bbc619412d5915c23507098df7014b4 --- /dev/null +++ b/physiological_integration/cases/case5/case5_21.json @@ -0,0 +1,95 @@ +{ + "meta_info": { + "case_id": "case05_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Extract Gaze point X (EyeTracking) and FP1 (EEG) from the same declared real-time interval [0, 1) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_21_signal_1.edf and the second at file_check/case05_21_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/SEEDVII_01_session-20221001_1.npy", + "data/multi/SEEDVII_01_session-20221001_1_continuous.npy", + "data/multi/SEEDVII_01_session-20221001_1_events.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_21_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 250.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Gaze point X" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1208.6647163046932, + "weight": 25, + "tolerance": 120.86647163046933 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_21_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "FP1" + ], + "weight": 25 + }, + "signal_rms": { + "value": 1.9476028621285715e-05, + "weight": 25, + "tolerance": 1.9476028621285718e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_22.json b/physiological_integration/cases/case5/case5_22.json new file mode 100644 index 0000000000000000000000000000000000000000..9735cca958a8e53f736cfd6883156c167c4c9a26 --- /dev/null +++ b/physiological_integration/cases/case5/case5_22.json @@ -0,0 +1,95 @@ +{ + "meta_info": { + "case_id": "case05_22", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_02_session-20221011_1_EEG.cnt", + "data/multi/SEEDVII_02_session-20221011_1_EyeTracking.edf" + ], + "instruction": "Extract HEO (EOG) and Pupil diameter right (EyeTracking) from the same declared real-time interval [4, 6) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_22_signal_1.edf and the second at file_check/case05_22_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/SEEDVII_02_session-20221011_1.npy", + "data/multi/SEEDVII_02_session-20221011_1_continuous.npy", + "data/multi/SEEDVII_02_session-20221011_1_events.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_22_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "HEO" + ], + "weight": 25 + }, + "signal_rms": { + "value": 0.0002238932345112451, + "weight": 25, + "tolerance": 2.2389323451124512e-05 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_22_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 250.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Pupil diameter r" + ], + "weight": 25 + }, + "signal_rms": { + "value": 4.763045043362272, + "weight": 25, + "tolerance": 0.47630450433622723 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_23.json b/physiological_integration/cases/case5/case5_23.json new file mode 100644 index 0000000000000000000000000000000000000000..83744d17a5eba3ca6840a8722b5deca5e33e2a46 --- /dev/null +++ b/physiological_integration/cases/case5/case5_23.json @@ -0,0 +1,95 @@ +{ + "meta_info": { + "case_id": "case05_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Extract ECG (ECG) and Gaze point Y (EyeTracking) from the same declared real-time interval [8, 11) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_23_signal_1.edf and the second at file_check/case05_23_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/SEEDVII_03_session-20221009_1.npy", + "data/multi/SEEDVII_03_session-20221009_1_continuous.npy", + "data/multi/SEEDVII_03_session-20221009_1_events.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_23_signal_1.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "ECG" + ], + "weight": 25 + }, + "signal_rms": { + "value": 9.280607252947558e-05, + "weight": 25, + "tolerance": 9.280607252947558e-06 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_23_signal_2.edf" + }, + "duration_sec": { + "value": 3.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 250.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Gaze point Y" + ], + "weight": 25 + }, + "signal_rms": { + "value": 791.338638012786, + "weight": 25, + "tolerance": 79.13386380127861 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_24.json b/physiological_integration/cases/case5/case5_24.json new file mode 100644 index 0000000000000000000000000000000000000000..d76394b5fd3e2085f943691cfe2896e24765b6cf --- /dev/null +++ b/physiological_integration/cases/case5/case5_24.json @@ -0,0 +1,95 @@ +{ + "meta_info": { + "case_id": "case05_24", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_04_session-20221030_1_EEG.cnt", + "data/multi/SEEDVII_04_session-20221030_1_EyeTracking.edf" + ], + "instruction": "Extract CZ (EEG) and Pupil diameter right (EyeTracking) from the same declared real-time interval [12, 13) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_24_signal_1.edf and the second at file_check/case05_24_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/SEEDVII_04_session-20221030_1.npy", + "data/multi/SEEDVII_04_session-20221030_1_continuous.npy", + "data/multi/SEEDVII_04_session-20221030_1_events.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_24_signal_1.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "CZ" + ], + "weight": 25 + }, + "signal_rms": { + "value": 4.180903641681288e-06, + "weight": 25, + "tolerance": 4.1809036416812884e-07 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_24_signal_2.edf" + }, + "duration_sec": { + "value": 1.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 250.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Pupil diameter r" + ], + "weight": 25 + }, + "signal_rms": { + "value": 6.0356729144672405, + "weight": 25, + "tolerance": 0.603567291446724 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case5/case5_25.json b/physiological_integration/cases/case5/case5_25.json new file mode 100644 index 0000000000000000000000000000000000000000..62517379f5ea8701935ea900f5529bcfcd1e82c3 --- /dev/null +++ b/physiological_integration/cases/case5/case5_25.json @@ -0,0 +1,95 @@ +{ + "meta_info": { + "case_id": "case05_25", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_05_session-20220929_1_EEG.cnt", + "data/multi/SEEDVII_05_session-20220929_1_EyeTracking.edf" + ], + "instruction": "Extract Gaze point X (EyeTracking) and VEO (EOG) from the same declared real-time interval [16, 18) seconds. Use each channel's native sampling rate and convert both time boundaries independently; never reuse sample indices between modalities. Apply the public annotation alignment rule and save each one-dimensional segment as a one-channel EDF: the first at file_check/case05_25_signal_1.edf and the second at file_check/case05_25_signal_2.edf. Return only signal_1_file and signal_2_file.", + "label_path": [ + "data/multi/SEEDVII_05_session-20220929_1.npy", + "data/multi/SEEDVII_05_session-20220929_1_continuous.npy", + "data/multi/SEEDVII_05_session-20220929_1_events.npy" + ] + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- signal_1_file: Signal 1 file.\n- signal_2_file: Signal 2 file.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"signal_1_file\": null, \"signal_2_file\": null}", + "metrics": [ + { + "metric_id": "signal_1_file_status", + "type": "file_status_check", + "target_key": "signal_1_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_25_signal_1.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 250.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "Gaze point X" + ], + "weight": 25 + }, + "signal_rms": { + "value": 765.7329274974448, + "weight": 25, + "tolerance": 76.57329274974448 + } + } + } + }, + { + "metric_id": "signal_2_file_status", + "type": "file_status_check", + "target_key": "signal_2_file", + "weight": 50, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "case05_25_signal_2.edf" + }, + "duration_sec": { + "value": 2.0, + "weight": 25, + "tolerance": 1e-06 + }, + "sfreq_hz": { + "value": 1000.0, + "weight": 25, + "tolerance": 1e-06 + }, + "channel_names": { + "value": [ + "VEO" + ], + "weight": 25 + }, + "signal_rms": { + "value": 3.1926646506221725e-05, + "weight": 25, + "tolerance": 3.1926646506221726e-06 + } + } + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_01.json b/physiological_integration/cases/case6/case6_01.json new file mode 100644 index 0000000000000000000000000000000000000000..6193eaf2dd405d78675be3b31b49ad60ee28220a --- /dev/null +++ b/physiological_integration/cases/case6/case6_01.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[160, 164)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[800, 804)`, `[1032, 1036)`, `[1264, 1268)`, `[1496, 1500)`, `[1728, 1732)`, `[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 2.893227024544046, + "tolerance": 0.2893227024544046 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 1.8144720834765704, + "tolerance": 0.18144720834765704 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_02.json b/physiological_integration/cases/case6/case6_02.json new file mode 100644 index 0000000000000000000000000000000000000000..0b7420e7d2c6161f0db28babbf6f991db8016e69 --- /dev/null +++ b/physiological_integration/cases/case6/case6_02.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[456, 460)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1032, 1036)`, `[1264, 1268)`, `[1496, 1500)`, `[1728, 1732)`, `[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.34210346404523095, + "tolerance": 0.0342103464045231 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.5779596343074518, + "tolerance": 0.05779596343074518 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_03.json b/physiological_integration/cases/case6/case6_03.json new file mode 100644 index 0000000000000000000000000000000000000000..f2715f4f193ec8bea9ec99757c02c91dac9fe791 --- /dev/null +++ b/physiological_integration/cases/case6/case6_03.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[752, 756)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1264, 1268)`, `[1496, 1500)`, `[1728, 1732)`, `[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 1.5600521510243661, + "tolerance": 0.15600521510243662 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 2.602274412011126, + "tolerance": 0.26022744120111263 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_04.json b/physiological_integration/cases/case6/case6_04.json new file mode 100644 index 0000000000000000000000000000000000000000..2b7c1036c00b091278606bae64ef8b52ac58e783 --- /dev/null +++ b/physiological_integration/cases/case6/case6_04.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[1048, 1052)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1496, 1500)`, `[1728, 1732)`, `[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -1.273156182428689, + "tolerance": 0.1273156182428689 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.8975184770575789, + "tolerance": 0.08975184770575789 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_05.json b/physiological_integration/cases/case6/case6_05.json new file mode 100644 index 0000000000000000000000000000000000000000..f01d05bbeb0a77de3c4111eb90c12dba2543e053 --- /dev/null +++ b/physiological_integration/cases/case6/case6_05.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[1344, 1348)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1728, 1732)`, `[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.2895303495731022, + "tolerance": 0.02895303495731022 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.06418491662537619, + "tolerance": 0.0064184916625376195 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_06.json b/physiological_integration/cases/case6/case6_06.json new file mode 100644 index 0000000000000000000000000000000000000000..ed30684507aaac40c506f7e1e2a87fe14a5cac50 --- /dev/null +++ b/physiological_integration/cases/case6/case6_06.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[1640, 1644)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1960, 1964)`, `[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 5.48499737334655, + "tolerance": 0.5484997373346551 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -1.2731301723671795, + "tolerance": 0.12731301723671795 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_07.json b/physiological_integration/cases/case6/case6_07.json new file mode 100644 index 0000000000000000000000000000000000000000..574e59323c605551e71db655795a013edfbc782a --- /dev/null +++ b/physiological_integration/cases/case6/case6_07.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[1936, 1940)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[2192, 2196)`, `[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.52037536267594, + "tolerance": 0.052037536267594 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.9313333761807303, + "tolerance": 0.09313333761807303 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_08.json b/physiological_integration/cases/case6/case6_08.json new file mode 100644 index 0000000000000000000000000000000000000000..74b7c30c06c5543b460aaf2e0d75643e47204926 --- /dev/null +++ b/physiological_integration/cases/case6/case6_08.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[2232, 2236)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[2424, 2428)`, `[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 0.37664327450595414, + "tolerance": 0.03766432745059542 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 1.3197687084017526, + "tolerance": 0.13197687084017526 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_09.json b/physiological_integration/cases/case6/case6_09.json new file mode 100644 index 0000000000000000000000000000000000000000..3615e457a6ce33932e3dfd3b867a7b30f6bf12ba --- /dev/null +++ b/physiological_integration/cases/case6/case6_09.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[2528, 2532)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[2656, 2660)`, `[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 0.028712821121909732, + "tolerance": 0.0028712821121909733 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 0.20322390410841817, + "tolerance": 0.020322390410841818 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_10.json b/physiological_integration/cases/case6/case6_10.json new file mode 100644 index 0000000000000000000000000000000000000000..8fb4a2c7318626d9b6826b17a9eb2d4d19636b46 --- /dev/null +++ b/physiological_integration/cases/case6/case6_10.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[2824, 2828)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[2888, 2892)`, `[3120, 3124)`, `[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.8010673402236365, + "tolerance": 0.08010673402236365 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.6568013715867304, + "tolerance": 0.06568013715867303 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_11.json b/physiological_integration/cases/case6/case6_11.json new file mode 100644 index 0000000000000000000000000000000000000000..65242fbe1cc2bc6e0f200820cf03b64ff9eabd42 --- /dev/null +++ b/physiological_integration/cases/case6/case6_11.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[3120, 3124)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.3821207633344238, + "tolerance": 0.03821207633344238 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 1.8554830453948326, + "tolerance": 0.18554830453948326 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_12.json b/physiological_integration/cases/case6/case6_12.json new file mode 100644 index 0000000000000000000000000000000000000000..22b0facedd21ee2a02d6d02d5a2d3313225bfb01 --- /dev/null +++ b/physiological_integration/cases/case6/case6_12.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[3416, 3420)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[3352, 3356)`, `[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.30873203737708627, + "tolerance": 0.03087320373770863 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 0.11354175605528183, + "tolerance": 0.011354175605528184 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_13.json b/physiological_integration/cases/case6/case6_13.json new file mode 100644 index 0000000000000000000000000000000000000000..c70b02e337d12b426ca4322141814fc8fffcae23 --- /dev/null +++ b/physiological_integration/cases/case6/case6_13.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[3712, 3716)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[3584, 3588)`, `[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 3.4180649876787865, + "tolerance": 0.34180649876787866 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 0.1643390047465754, + "tolerance": 0.01643390047465754 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_14.json b/physiological_integration/cases/case6/case6_14.json new file mode 100644 index 0000000000000000000000000000000000000000..2d628b0fb4832ee9ddd70f33008a03936f6ae8c5 --- /dev/null +++ b/physiological_integration/cases/case6/case6_14.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[4008, 4012)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[3816, 3820)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 0.40799902517710057, + "tolerance": 0.04079990251771006 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 0.25909978261686817, + "tolerance": 0.025909978261686818 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_15.json b/physiological_integration/cases/case6/case6_15.json new file mode 100644 index 0000000000000000000000000000000000000000..809fd7aaaacdb2987ece105382475c38ff906732 --- /dev/null +++ b/physiological_integration/cases/case6/case6_15.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[4304, 4308)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[4048, 4052)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.7394781523643789, + "tolerance": 0.07394781523643788 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -1.7596208675962774, + "tolerance": 0.17596208675962777 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_16.json b/physiological_integration/cases/case6/case6_16.json new file mode 100644 index 0000000000000000000000000000000000000000..44b186a2a9d8fcff9c19c3c085c1d0c34038d7b5 --- /dev/null +++ b/physiological_integration/cases/case6/case6_16.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_16", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[4600, 4604)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[1232, 1236)`, `[4280, 4284)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 0.5922913243242521, + "tolerance": 0.05922913243242522 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": 1.6079709770929234, + "tolerance": 0.16079709770929235 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_17.json b/physiological_integration/cases/case6/case6_17.json new file mode 100644 index 0000000000000000000000000000000000000000..9f176125fd37696c385cdd9eee076ad045858836 --- /dev/null +++ b/physiological_integration/cases/case6/case6_17.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[4896, 4900)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[1232, 1236)`, `[1464, 1468)`, `[4512, 4516)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.024606365971660077, + "tolerance": 0.002460636597166008 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.17885934766939568, + "tolerance": 0.01788593476693957 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_18.json b/physiological_integration/cases/case6/case6_18.json new file mode 100644 index 0000000000000000000000000000000000000000..1c045a6f4bfc7bbadb58bbd74b1c6c40dda050da --- /dev/null +++ b/physiological_integration/cases/case6/case6_18.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_18", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[5192, 5196)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[1232, 1236)`, `[1464, 1468)`, `[1696, 1700)`, `[4744, 4748)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": 2.5307747744050615, + "tolerance": 0.2530774774405062 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.24909778019896725, + "tolerance": 0.024909778019896727 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_19.json b/physiological_integration/cases/case6/case6_19.json new file mode 100644 index 0000000000000000000000000000000000000000..efd99dcdcf29f767e1273bd7ae60fa3d5bff1e31 --- /dev/null +++ b/physiological_integration/cases/case6/case6_19.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[5488, 5492)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[1232, 1236)`, `[1464, 1468)`, `[1696, 1700)`, `[1928, 1932)`, `[4976, 4980)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.0622272576526257, + "tolerance": 0.0062227257652625705 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.631393592971362, + "tolerance": 0.0631393592971362 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case6/case6_20.json b/physiological_integration/cases/case6/case6_20.json new file mode 100644 index 0000000000000000000000000000000000000000..8a6c2e9cfb504e4f863d4ea9a18946231c3bbde3 --- /dev/null +++ b/physiological_integration/cases/case6/case6_20.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "case06_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Using EEG channels `FT7, FT8, T7, T8, TP7, TP8` and EOG channels `EOG-H, EOG-V`, analyze the target interval `[5784, 5788)` seconds.\n\nFor the target interval and each of the following same-record reference intervals—`[1000, 1004)`, `[1232, 1236)`, `[1464, 1468)`, `[1696, 1700)`, `[1928, 1932)`, `[2160, 2164)`, `[5208, 5212)`, `[5440, 5444)`, `[5672, 5676)`, `[5904, 5908)`, `[6136, 6140)`, `[6368, 6372)` seconds—compute the RMS of each listed channel using its finite samples, then take the median across channels separately for EEG and EOG. Do not apply a logarithmic transform.\n\nFor each modality, standardize the target median-RMS value using the mean and population standard deviation (ddof=0) of the corresponding 12 reference median-RMS values. The reference standard deviation is guaranteed to be positive.\n\nSet `artifact_present` to `true` only if both `eeg_z_score` and `eog_z_score` are at least 1.0; otherwise, set it to `false`.\n\nReturn only one valid JSON object containing exactly:\n\n{\n\"eeg_z_score\": , \n\"eog_z_score\": ,\n\"artifact_present\": \n}" + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- eeg_z_score: Eeg z score. A finite z-score for the EEG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EEG reference median-RMS values.\n- eog_z_score: Eog z score. A finite z-score for the EOG target median-RMS value, standardized with the mean and population standard deviation (ddof=0) of the 12 EOG reference median-RMS values.\n- artifact_present: Artifact present. true only when both eeg_z_score and eog_z_score are at least 1.0; otherwise false.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"eeg_z_score\": , \"eog_z_score\": , \"artifact_present\": }", + "metrics": [ + { + "metric_id": "eeg_z_score", + "type": "numeric_check", + "target_key": "eeg_z_score", + "weight": 40, + "params": { + "gt_value": -0.799956482042844, + "tolerance": 0.0799956482042844 + } + }, + { + "metric_id": "eog_z_score", + "type": "numeric_check", + "target_key": "eog_z_score", + "weight": 40, + "params": { + "gt_value": -0.5408570114570364, + "tolerance": 0.05408570114570364 + } + }, + { + "metric_id": "artifact_present", + "type": "categorical_check", + "target_key": "artifact_present", + "weight": 20, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_01.json b/physiological_integration/cases/case7/case7_01.json new file mode 100644 index 0000000000000000000000000000000000000000..67ca358548387f7fb2ccf49389c051985bb5c646 --- /dev/null +++ b/physiological_integration/cases/case7/case7_01.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "Analyze the half-open interval `[132.072266, 192.201172)` seconds in the DEAP recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. For DEAP, treat EXG1 and EXG2 as EOG and EXG3 and EXG4 as EMG; exclude EXG5 through EXG8 and Erg1 and Erg2. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "PLET", + "FC1", + "FC2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_02.json b/physiological_integration/cases/case7/case7_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b7a26ba51d9f20b87ae3b9c9bf845d912f9a39f5 --- /dev/null +++ b/physiological_integration/cases/case7/case7_02.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "Analyze the half-open interval `[979.412109, 1039.490234)` seconds in the DEAP recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. For DEAP, treat EXG1 and EXG2 as EOG and EXG3 and EXG4 as EMG; exclude EXG5 through EXG8 and Erg1 and Erg2. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "TEMP", + "FP2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_03.json b/physiological_integration/cases/case7/case7_03.json new file mode 100644 index 0000000000000000000000000000000000000000..cf9b5a08f0c1f5d8df4f29962f122bf1f38935bf --- /dev/null +++ b/physiological_integration/cases/case7/case7_03.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "Analyze the half-open interval `[201.345703, 261.423828)` seconds in the DEAP recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. For DEAP, treat EXG1 and EXG2 as EOG and EXG3 and EXG4 as EMG; exclude EXG5 through EXG8 and Erg1 and Erg2. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "PLET", + "CZ", + "FP1" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_04.json b/physiological_integration/cases/case7/case7_04.json new file mode 100644 index 0000000000000000000000000000000000000000..02ae7cc237a82ff35ae448795f5adef5df684ba2 --- /dev/null +++ b/physiological_integration/cases/case7/case7_04.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "Analyze the half-open interval `[159.484375, 219.564453)` seconds in the DEAP recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. For DEAP, treat EXG1 and EXG2 as EOG and EXG3 and EXG4 as EMG; exclude EXG5 through EXG8 and Erg1 and Erg2. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "RESP", + "PLET", + "PO3" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_05.json b/physiological_integration/cases/case7/case7_05.json new file mode 100644 index 0000000000000000000000000000000000000000..832e87839dcf31090fdc768f2f8c6714a1bc5cfb --- /dev/null +++ b/physiological_integration/cases/case7/case7_05.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "Analyze the half-open interval `[372.791016, 432.851562)` seconds in the DEAP recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. For DEAP, treat EXG1 and EXG2 as EOG and EXG3 and EXG4 as EMG; exclude EXG5 through EXG8 and Erg1 and Erg2. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EXG2", + "RESP", + "FC5" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_06.json b/physiological_integration/cases/case7/case7_06.json new file mode 100644 index 0000000000000000000000000000000000000000..c0328888c5e33ddc864e01d1a547c6ea475cf748 --- /dev/null +++ b/physiological_integration/cases/case7/case7_06.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "Analyze the half-open interval `[8, 16)` seconds in the SEED-VIG recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG-V", + "TP8", + "EOG-H" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_07.json b/physiological_integration/cases/case7/case7_07.json new file mode 100644 index 0000000000000000000000000000000000000000..c32953e6bc9bcabbc864cdad664238679373487a --- /dev/null +++ b/physiological_integration/cases/case7/case7_07.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "Analyze the half-open interval `[96, 104)` seconds in the SEED-VIG recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG-V", + "EOG-H", + "FT8" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_08.json b/physiological_integration/cases/case7/case7_08.json new file mode 100644 index 0000000000000000000000000000000000000000..f0051fff86c19a50d31bcb16235821d54c7baafd --- /dev/null +++ b/physiological_integration/cases/case7/case7_08.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "Analyze the half-open interval `[24, 32)` seconds in the SEED-VIG recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG-H", + "T7", + "TP7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_09.json b/physiological_integration/cases/case7/case7_09.json new file mode 100644 index 0000000000000000000000000000000000000000..51240c3942bf1f736edf3f14e1a0cea43e453454 --- /dev/null +++ b/physiological_integration/cases/case7/case7_09.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "Analyze the half-open interval `[72, 80)` seconds in the SEED-VIG recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG-V", + "CP1", + "EOG-H" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_10.json b/physiological_integration/cases/case7/case7_10.json new file mode 100644 index 0000000000000000000000000000000000000000..7ab1f021cfa9116c9034ae0c5ce3de4a098acc9d --- /dev/null +++ b/physiological_integration/cases/case7/case7_10.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "Analyze the half-open interval `[32, 40)` seconds in the SEED-VIG recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG-V", + "EOG-H", + "T8" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_11.json b/physiological_integration/cases/case7/case7_11.json new file mode 100644 index 0000000000000000000000000000000000000000..5209cc345228c63c368379f2962ca962e18ecbb4 --- /dev/null +++ b/physiological_integration/cases/case7/case7_11.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_11", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_01.edf", + "instruction": "Analyze the half-open interval `[0, 30)` seconds in the SHHS1 recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "ABDO RES", + "THOR RES", + "EMG" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_12.json b/physiological_integration/cases/case7/case7_12.json new file mode 100644 index 0000000000000000000000000000000000000000..ed2297fd05142e371671d71b79c5a90cb8284812 --- /dev/null +++ b/physiological_integration/cases/case7/case7_12.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_12", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_02.edf", + "instruction": "Analyze the half-open interval `[0, 30)` seconds in the SHHS1 recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "THOR RES", + "EEG", + "ABDO RES" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_13.json b/physiological_integration/cases/case7/case7_13.json new file mode 100644 index 0000000000000000000000000000000000000000..f70a3ab9f61502cc211560af368fa30743eec0cc --- /dev/null +++ b/physiological_integration/cases/case7/case7_13.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_13", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_03.edf", + "instruction": "Analyze the half-open interval `[0, 30)` seconds in the SHHS1 recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "ECG", + "EMG", + "EOG(L)" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_14.json b/physiological_integration/cases/case7/case7_14.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f35d71f2933263ae4149671fa76ea4c5b18b1c --- /dev/null +++ b/physiological_integration/cases/case7/case7_14.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_14", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_04.edf", + "instruction": "Analyze the half-open interval `[0, 30)` seconds in the SHHS1 recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EOG(L)", + "EMG", + "ECG" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_15.json b/physiological_integration/cases/case7/case7_15.json new file mode 100644 index 0000000000000000000000000000000000000000..a7d829c8a4697ba1dbaad8aca8c93ddbbc3f1d63 --- /dev/null +++ b/physiological_integration/cases/case7/case7_15.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "case07_15", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SHHS1" + }, + "agent_input": { + "data_path": "data/multi/SHHS1_05.edf", + "instruction": "Analyze the half-open interval `[150, 180)` seconds in the SHHS1 recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "EMG", + "EOG(L)", + "EOG(R)" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_17.json b/physiological_integration/cases/case7/case7_17.json new file mode 100644 index 0000000000000000000000000000000000000000..15dc2e48748dea51fe9cbd9390e866d5e1d9187b --- /dev/null +++ b/physiological_integration/cases/case7/case7_17.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case07_17", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_02_run-gonogo1_EEG.edf", + "data/multi/simultaneous_02_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze the half-open interval `[113.026, 182.136)` seconds in the simultaneous recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "HEOG", + "S05D05_850", + "VEOG" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_19.json b/physiological_integration/cases/case7/case7_19.json new file mode 100644 index 0000000000000000000000000000000000000000..74bd815e4171eb270ef67ea8e062a85e3e0c5b3b --- /dev/null +++ b/physiological_integration/cases/case7/case7_19.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "case07_19", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_04_run-gonogo1_EEG.edf", + "data/multi/simultaneous_04_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze the half-open interval `[177.37, 245.18)` seconds in the simultaneous recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the four channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the four channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final four highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than four channels are discussed, extract only those explicitly identified as the final top four.\n14. If the final set is missing, ambiguous, or contains fewer than four identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "HEOG", + "FP2", + "VEOG" + ], + "match_mode": "per_hit_fraction", + "top_k": 4, + "score_per_hit": 0.25 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_20.json b/physiological_integration/cases/case7/case7_20.json new file mode 100644 index 0000000000000000000000000000000000000000..1fe92a21f72bdd3859bee0554a369649b4b0a500 --- /dev/null +++ b/physiological_integration/cases/case7/case7_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case07_20", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "simultaneous" + }, + "agent_input": { + "data_path": [ + "data/multi/simultaneous_05_run-gonogo1_EEG.edf", + "data/multi/simultaneous_05_run-gonogo1_fNIRS.edf" + ], + "instruction": "Analyze the half-open interval `[332.247, 399.657)` seconds in the simultaneous recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "VEOG", + "FP2", + "FP1" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_21.json b/physiological_integration/cases/case7/case7_21.json new file mode 100644 index 0000000000000000000000000000000000000000..4a408a1c8f173c32fac601b45e05c37edc4b3ec6 --- /dev/null +++ b/physiological_integration/cases/case7/case7_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "case07_21", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_01_session-20221001_1_EEG.cnt", + "data/multi/SEEDVII_01_session-20221001_1_EyeTracking.edf" + ], + "instruction": "Analyze the half-open interval `[2878.414, 3097.872)` seconds in the SEED-VII recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the two channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the two channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final two highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than two channels are discussed, extract only those explicitly identified as the final top two.\n14. If the final set is missing, ambiguous, or contains fewer than two identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "ECG", + "M2" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} diff --git a/physiological_integration/cases/case7/case7_23.json b/physiological_integration/cases/case7/case7_23.json new file mode 100644 index 0000000000000000000000000000000000000000..4418f16e247433dc242a8964f0737d98df49e712 --- /dev/null +++ b/physiological_integration/cases/case7/case7_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "case07_23", + "bench_subset": "NeuroBench-Multi", + "difficult": 1.5, + "original_dataset": "SEED-VII" + }, + "agent_input": { + "data_path": [ + "data/multi/SEEDVII_03_session-20221009_1_EEG.cnt", + "data/multi/SEEDVII_03_session-20221009_1_EyeTracking.edf" + ], + "instruction": "Analyze the half-open interval `[3030.329, 3177.8)` seconds in the SEED-VII recording, using each channel at its native sampling rate.\n\nUse only analyzable signal channels representing EEG, EOG, EMG, ECG, GSR/EDA, respiration or airflow, plethysmography or pulse, temperature or oximetry, fNIRS intensity, gaze position, or pupil size when present. Exclude Status, event/marker/trigger fields, validity channels, position and light metadata, and unlabeled auxiliary acquisition channels. Retain only channels with at least two finite full-record samples and a finite positive scaled MAD.\n\nIndependently normalize every retained channel using all of its finite samples over its complete recording: subtract the full-record median and divide by the scaled MAD, defined as `1.4826 * median(abs(x - median(x)))`. Then extract that channel's samples in the specified interval and compute their population standard deviation (`ddof=0`) using only finite normalized samples. Use this standard deviation as the channel's variability score.\n\nRank all retained channels by this score and return `highest_volatility_channels` as the unordered set of the three channels with the largest scores." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multimodal physiological data. Your sole task is to convert the agent's report into one structured JSON object.\n\n### TASK\nExtract the final set of the three channels reported as having the highest signal variability in the specified time interval.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text or explanations.\n4. The key must be EXACTLY \"highest_volatility_channels\".\n5. The value must be an array of channel-name strings.\n6. Extract only the final three highest-volatility channels. Do not include intermediate rankings, all analyzed channels, excluded channels, or channels mentioned only as examples.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy EEG aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, and T6->P8.\n10. Preserve non-EEG physiological channel labels after trimming and uppercasing, such as GSR1, RESP, PLET, and TEMP.\n11. Remove duplicate channel labels.\n12. The channel order has no meaning, but preserve the final reported order when possible.\n13. If more than three channels are discussed, extract only those explicitly identified as the final top three.\n14. If the final set is missing, ambiguous, or contains fewer than three identifiable channels, return {\"highest_volatility_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"highest_volatility_channels\": [\"CH1\", \"CH2\", \"CH3\"]}", + "metrics": [ + { + "metric_id": "highest_volatility_channel_set", + "type": "set_match_check", + "target_key": "highest_volatility_channels", + "weight": 100, + "params": { + "gt_value": [ + "M1", + "M2", + "ECG" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_01.json b/physiological_integration/cases/case8/case8_01.json new file mode 100644 index 0000000000000000000000000000000000000000..086339fcd63791daa4e214c30ca45b7db187db2b --- /dev/null +++ b/physiological_integration/cases/case8/case8_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "From GSR/EDA channel `GSR1`, analyze the half-open interval `[0, 60)` seconds at its native sampling rate.\n\nApply the declared fourth-order 5 Hz low-pass filter.\nEstimate the tonic component using a centered 5120-sample moving average with reflect padding, and define the phasic component as the filtered signal minus the tonic component.\nSet `feature_value` to the mean of the tonic component over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 5 Hz low-pass filter to channel `GSR1` over `[0, 60)`. Estimate the tonic component with a centered 5120-sample moving average using reflect padding, and define the phasic component as filtered signal minus tonic component. Report the final requested scalar: the mean of the tonic component.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 5130.347796, + "tolerance": 513.0347796 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_02.json b/physiological_integration/cases/case8/case8_02.json new file mode 100644 index 0000000000000000000000000000000000000000..cfccad37afc08bc6a33f7eed790729d6500e729e --- /dev/null +++ b/physiological_integration/cases/case8/case8_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "From GSR/EDA channel `GSR1`, analyze the half-open interval `[60, 120)` seconds at its native sampling rate.\n\nApply the declared fourth-order 5 Hz low-pass filter.\nEstimate the tonic component using a centered 5120-sample moving average with reflect padding, and define the phasic component as the filtered signal minus the tonic component.\nSet `feature_value` to the linear slope per second of the tonic component over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 5 Hz low-pass filter to channel `GSR1` over `[60, 120)`. Estimate the tonic component with a centered 5120-sample moving average using reflect padding, and define the phasic component as filtered signal minus tonic component. Report the final requested scalar: the linear slope per second of the tonic component.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 0.310323, + "tolerance": 0.031032300000000002 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_03.json b/physiological_integration/cases/case8/case8_03.json new file mode 100644 index 0000000000000000000000000000000000000000..e98fc6190ea8e46cdf738b096dbccab5ec2f6057 --- /dev/null +++ b/physiological_integration/cases/case8/case8_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "From GSR/EDA channel `GSR1`, analyze the half-open interval `[120, 180)` seconds at its native sampling rate.\n\nApply the declared fourth-order 5 Hz low-pass filter.\nEstimate the tonic component using a centered 5120-sample moving average with reflect padding, and define the phasic component as the filtered signal minus the tonic component.\nDetect peaks in the phasic component using a minimum distance of 512 samples and prominence max(0.01, 0.5 × std(phasic)).\nSet `feature_value` to the number of peaks in the phasic component over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 5 Hz low-pass filter to channel `GSR1` over `[120, 180)`. Estimate the tonic component with a centered 5120-sample moving average using reflect padding, and define the phasic component as filtered signal minus tonic component. Detect phasic peaks with a minimum distance of 512 samples and prominence max(0.01, 0.5 × std(phasic)). Report the final requested scalar: the number of peaks in the phasic component.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 6.0, + "tolerance": 0.6000000000000001 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_04.json b/physiological_integration/cases/case8/case8_04.json new file mode 100644 index 0000000000000000000000000000000000000000..9008c738bf05452f454d45c33ac614200eaa8061 --- /dev/null +++ b/physiological_integration/cases/case8/case8_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "From GSR/EDA channel `GSR1`, analyze the half-open interval `[180, 240)` seconds at its native sampling rate.\n\nApply the declared fourth-order 5 Hz low-pass filter.\nEstimate the tonic component using a centered 5120-sample moving average with reflect padding, and define the phasic component as the filtered signal minus the tonic component.\nSet `feature_value` to the positive trapezoid area of the phasic component over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 5 Hz low-pass filter to channel `GSR1` over `[180, 240)`. Estimate the tonic component with a centered 5120-sample moving average using reflect padding, and define the phasic component as filtered signal minus tonic component. Report the final requested scalar: the positive trapezoid area of the phasic component.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 23.891037, + "tolerance": 2.3891037 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_05.json b/physiological_integration/cases/case8/case8_05.json new file mode 100644 index 0000000000000000000000000000000000000000..4f52970ecb627095bb59cb5bdd607c0e1bb6b0f2 --- /dev/null +++ b/physiological_integration/cases/case8/case8_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "From Respiration channel `Resp`, analyze the half-open interval `[0, 60)` seconds at its native sampling rate.\n\nApply the declared fourth-order 0.05–1 Hz band-pass filter.\nRobust-z-score the filtered signal to obtain the normalized signal.\nDetect peaks using a minimum distance of 1024 samples and prominence 0.5.\nSet `feature_value` to the mean peak-interval rate per minute of the normalized signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 0.05–1 Hz band-pass filter to channel `Resp` over `[0, 60)`, robust-z-score the filtered signal, and detect peaks with a minimum distance of 1024 samples and prominence 0.5. Report the final requested scalar: the mean peak-interval rate per minute.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 17.973756, + "tolerance": 1.7973756000000003 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_06.json b/physiological_integration/cases/case8/case8_06.json new file mode 100644 index 0000000000000000000000000000000000000000..8e036f7f0de9ea2cbd6cfd32b74f541dfc42c03b --- /dev/null +++ b/physiological_integration/cases/case8/case8_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_01.bdf", + "instruction": "From Respiration channel `Resp`, analyze the half-open interval `[60, 120)` seconds at its native sampling rate.\n\nApply the declared fourth-order 0.05–1 Hz band-pass filter.\nRobust-z-score the filtered signal to obtain the normalized signal.\nDetect peaks using a minimum distance of 1024 samples and prominence 0.5.\nSet `feature_value` to the mean peak-to-preceding-trough amplitude of the normalized signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 0.05–1 Hz band-pass filter to channel `Resp` over `[60, 120)`, robust-z-score the filtered signal, and detect peaks with a minimum distance of 1024 samples and prominence 0.5. Report the final requested scalar: the mean peak-to-preceding-trough amplitude.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 2.359312025004674, + "tolerance": 0.2359312025004674 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_07.json b/physiological_integration/cases/case8/case8_07.json new file mode 100644 index 0000000000000000000000000000000000000000..67d9fafda33e989fbfe470ead280ea42f2936cc3 --- /dev/null +++ b/physiological_integration/cases/case8/case8_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_02.bdf", + "instruction": "From PPG/BVP channel `Plet`, analyze the half-open interval `[0, 60)` seconds at its native sampling rate.\n\nApply the declared fourth-order 0.5–8 Hz band-pass filter.\nRobust-z-score the filtered signal to obtain the normalized signal.\nDetect peaks using a minimum distance of 205 samples and prominence 0.5.\nSet `feature_value` to the mean peak-interval rate per minute of the normalized signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 0.5–8 Hz band-pass filter to channel `Plet` over `[0, 60)`, robust-z-score the filtered signal, and detect peaks with a minimum distance of 205 samples and prominence 0.5. Report the final requested scalar: the mean peak-interval rate per minute.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 68.982596, + "tolerance": 6.8982596 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_08.json b/physiological_integration/cases/case8/case8_08.json new file mode 100644 index 0000000000000000000000000000000000000000..45c8b67e257499b0facb8fdbb0e180bf7b822148 --- /dev/null +++ b/physiological_integration/cases/case8/case8_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_03.bdf", + "instruction": "From PPG/BVP channel `Plet`, analyze the half-open interval `[60, 120)` seconds at its native sampling rate.\n\nApply the declared fourth-order 0.5–8 Hz band-pass filter.\nRobust-z-score the filtered signal to obtain the normalized signal.\nDetect peaks using a minimum distance of 205 samples and prominence 0.5.\nSet `feature_value` to the mean peak-to-preceding-trough amplitude of the normalized signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nApply the declared fourth-order 0.5–8 Hz band-pass filter to channel `Plet` over `[60, 120)`, robust-z-score the filtered signal, and detect peaks with a minimum distance of 205 samples and prominence 0.5. Report the final requested scalar: the mean peak-to-preceding-trough amplitude.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 3.454834170463396, + "tolerance": 0.3454834170463396 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_09.json b/physiological_integration/cases/case8/case8_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e09c0a1c535c98b3108ed3170a2f24d467e44c9f --- /dev/null +++ b/physiological_integration/cases/case8/case8_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_04.bdf", + "instruction": "From Temperature channel `Temp`, analyze the half-open interval `[0, 60)` seconds at its native sampling rate.\n\nUse the raw signal without an additional filter.\nSmooth it with a centered 2560-sample moving average with reflect padding.\nSet `feature_value` to the mean of the smoothed signal over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nUse channel `Temp` over `[0, 60)` without an additional filter. Apply a centered 2560-sample moving average with reflect padding and report the final requested scalar: the mean of the smoothed signal.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": 32.658024, + "tolerance": 3.2658024 + } + } + ] + } +} diff --git a/physiological_integration/cases/case8/case8_10.json b/physiological_integration/cases/case8/case8_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5b491a3960bcdf2e36f73684e50f882cb69c27e7 --- /dev/null +++ b/physiological_integration/cases/case8/case8_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case08_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "DEAP" + }, + "agent_input": { + "data_path": "data/multi/DEAP_05.bdf", + "instruction": "From Temperature channel `Temp`, analyze the half-open interval `[60, 180)` seconds at its native sampling rate.\n\nUse the raw signal without an additional filter.\nSmooth it with a centered 2560-sample moving average with reflect padding.\nSet `feature_value` to the linear slope per second of the smoothed signal over the interval." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final scalar feature produced by the complete declared signal-analysis procedure.\nUse channel `Temp` over `[60, 180)` without an additional filter. Apply a centered 2560-sample moving average with reflect padding and report the final requested scalar: the linear slope per second of the smoothed signal.\nExtract the final requested aggregate only, not intermediate filtered, tonic, phasic, normalized, peak, or smoothed values.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The JSON key must be EXACTLY `feature_value`.\n5. The value must be one finite numeric scalar; integer counts are allowed when the requested feature is a count.\n6. If the final feature value is missing or cannot be identified, return {\"feature_value\": null}.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100.0, + "params": { + "gt_value": -0.00976, + "tolerance": 0.000976 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_01.json b/physiological_integration/cases/case9/case9_01.json new file mode 100644 index 0000000000000000000000000000000000000000..bbc2ffd1ef13c6d576666935e75a952e219849c5 --- /dev/null +++ b/physiological_integration/cases/case9/case9_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_01", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 139.0810619069781, + "tolerance": 13.908106190697811 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_02.json b/physiological_integration/cases/case9/case9_02.json new file mode 100644 index 0000000000000000000000000000000000000000..17f0f1955cbb10db2ddc7573682d9fedd69fd683 --- /dev/null +++ b/physiological_integration/cases/case9/case9_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_02", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 61.05082239310536, + "tolerance": 6.105082239310537 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_03.json b/physiological_integration/cases/case9/case9_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c3941100ff39c0194059c15e68f914e757a955ef --- /dev/null +++ b/physiological_integration/cases/case9/case9_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_03", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 100.93785540516382, + "tolerance": 10.093785540516382 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_04.json b/physiological_integration/cases/case9/case9_04.json new file mode 100644 index 0000000000000000000000000000000000000000..e5734118fc59110d7fb8949014b717774e91ffd7 --- /dev/null +++ b/physiological_integration/cases/case9/case9_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_04", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 178.921019379727, + "tolerance": 17.8921019379727 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_05.json b/physiological_integration/cases/case9/case9_05.json new file mode 100644 index 0000000000000000000000000000000000000000..d0e6e9470aefb30d22cd11091830248c0715d687 --- /dev/null +++ b/physiological_integration/cases/case9/case9_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_05", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 74.51237825007526, + "tolerance": 7.451237825007526 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_06.json b/physiological_integration/cases/case9/case9_06.json new file mode 100644 index 0000000000000000000000000000000000000000..40b25a688862b9eef6174819c60a3aac6825f44e --- /dev/null +++ b/physiological_integration/cases/case9/case9_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_06", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_01_session-20151124_noon_2.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 139.0810619069781, + "tolerance": 13.908106190697811 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_07.json b/physiological_integration/cases/case9/case9_07.json new file mode 100644 index 0000000000000000000000000000000000000000..2b09d6d5005a47eb92ffd212f3af2ea5f4ca9c75 --- /dev/null +++ b/physiological_integration/cases/case9/case9_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_07", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_02_session-20151106_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 61.05082239310536, + "tolerance": 6.105082239310537 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_08.json b/physiological_integration/cases/case9/case9_08.json new file mode 100644 index 0000000000000000000000000000000000000000..3eab979bdd1c3781c444e2590c34000df8a2ce16 --- /dev/null +++ b/physiological_integration/cases/case9/case9_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_08", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_03_session-20151024_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 100.93785540516382, + "tolerance": 10.093785540516382 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_09.json b/physiological_integration/cases/case9/case9_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e237961220adeaa4443c0c84aab492c810f5bf8f --- /dev/null +++ b/physiological_integration/cases/case9/case9_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_09", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_04_session-20151105_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 178.921019379727, + "tolerance": 17.8921019379727 + } + } + ] + } +} diff --git a/physiological_integration/cases/case9/case9_10.json b/physiological_integration/cases/case9/case9_10.json new file mode 100644 index 0000000000000000000000000000000000000000..0be5de271a74c376f53a1efd343a2b3a5918bc19 --- /dev/null +++ b/physiological_integration/cases/case9/case9_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "case09_10", + "bench_subset": "NeuroBench-Multi", + "difficult": 1, + "original_dataset": "SEED-VIG" + }, + "agent_input": { + "data_path": "data/multi/SEEDVIG_05_session-20141108_noon.edf", + "instruction": "From EOG channel `eog_v`, compute the EOG RMS over the half-open interval `[80, 88)` seconds.\n\nUsing the channel’s native sampling rate and physical values decoded from the EDF file, convert them to microvolts (uV). Apply a fourth-order zero-phase Butterworth band-pass filter from `0.1` to `15` Hz to the full available channel. Then extract `[80, 88)` seconds from the filtered signal and compute its root-mean-square value in uV.\n\nReport only `feature_value`." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the following final answer fields from the agent's report into one JSON object:\n- feature_value: Feature value.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or extra fields.\n4. The JSON keys must be EXACTLY the keys listed in the OUTPUT TEMPLATE.\n5. Preserve the declared value type for every key.\n6. If a requested value is missing or cannot be found, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"feature_value\": }", + "metrics": [ + { + "metric_id": "feature_value_accuracy", + "type": "numeric_check", + "target_key": "feature_value", + "weight": 100, + "params": { + "gt_value": 74.51237825007526, + "tolerance": 7.451237825007526 + } + } + ] + } +}