{ "meta_info": { "case_id": "HMC_04.edf", "bench_subset": "NeuroBench-Sleep", "difficult": 1.5, "original_dataset": "HMC" }, "agent_input": { "data_path": "data/sleep/HMC_04.edf", "instruction": "Please analyze the EEG signals in the following three 5-minute windows of the provided sleep file: A: 496.5-501.5 minutes; B: 379.5-384.5 minutes; C: 43.0-48.0 minutes. Select EEG channels only. Determine which window is most slow-wave-rich, and report the delta-band relative power percentage for EEG channels in each of the three windows. Define delta relative power as power in 0.5-4Hz divided by total power in 0.5-40Hz, multiplied by 100. Report the selected window using only A, B, or C, and clearly state the delta relative power percentages for A, B, and C." }, "eval_config": { "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep EEG analysis reports. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract four final outputs from the agent's report:\n1) which one of windows A, B, and C is most slow-wave-rich\n2) the delta relative power percentage for window A\n3) the delta relative power percentage for window B\n4) the delta relative power percentage for window C\n\n### STRICT 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 keys must be EXACTLY \"slow_wave_rich_window\", \"delta_ratio_A\", \"delta_ratio_B\", and \"delta_ratio_C\".\n5. \"slow_wave_rich_window\" must be exactly one of: \"A\", \"B\", \"C\", or null.\n6. Each delta ratio value must be a FLOAT percentage without a percent sign.\n7. If a report gives a fraction or proportion between 0 and 1, convert it to a percentage by multiplying by 100.\n8. If a value is missing or unclear, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"slow_wave_rich_window\": <\"A\"|\"B\"|\"C\"|null>, \"delta_ratio_A\": , \"delta_ratio_B\": , \"delta_ratio_C\": }", "metrics": [ { "metric_id": "slow_wave_rich_window_accuracy", "type": "categorical_check", "target_key": "slow_wave_rich_window", "weight": 40, "params": { "gt_value": "C" } }, { "metric_id": "delta_ratio_A_accuracy", "type": "numeric_check", "target_key": "delta_ratio_A", "weight": 20, "params": { "gt_value": 46.901744669264325, "tolerance": 2.3450872334632162 } }, { "metric_id": "delta_ratio_B_accuracy", "type": "numeric_check", "target_key": "delta_ratio_B", "weight": 20, "params": { "gt_value": 60.26390585909406, "tolerance": 3.0131952929547032 } }, { "metric_id": "delta_ratio_C_accuracy", "type": "numeric_check", "target_key": "delta_ratio_C", "weight": 20, "params": { "gt_value": 84.39028196005106, "tolerance": 4.219514098002553 } } ] } }