| { |
| "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\": \"<early|middle|late|null>\", \"eog_peak_phase\": \"<early|middle|late|null>\"}", |
| "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" |
| } |
| } |
| ] |
| } |
| } |
|
|
