{
  "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\":\"<A|B|C|null>\",\"supporting_modality\":\"<EOG|EMG|neither|null>\"}",
    "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"
        }
      }
    ]
  }
}