Datasets:
| { | |
| "meta_info": { | |
| "case_id": "ISRUC_02.edf", | |
| "bench_subset": "NeuroBench-Sleep", | |
| "difficult": 1, | |
| "original_dataset": "ISRUC" | |
| }, | |
| "agent_input": { | |
| "label_path": "data/sleep/ISRUC_02.npy", | |
| "instruction": "Please use the provided sleep-stage label file to calculate two transition-based metrics for the whole night: the total number of sleep-stage transitions, and the number of interruptions from deep sleep into wakefulness. Please clearly report both final counts." | |
| }, | |
| "eval_config": { | |
| "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep medicine reports. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two transition counts from the agent's report:\n1) the total number of sleep-stage transitions across the whole night\n2) the number of deep-sleep interruptions into wakefulness\n\n### STRICT 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 \"total_stage_transitions\" and \"n3_to_w_interruptions\".\n5. Both values must be INTEGERS or null.\n6. If one value is missing or cannot be found, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"total_stage_transitions\": <integer|null>, \"n3_to_w_interruptions\": <integer|null>}", | |
| "metrics": [ | |
| { | |
| "metric_id": "total_stage_transition_accuracy", | |
| "type": "numeric_check", | |
| "target_key": "total_stage_transitions", | |
| "weight": 50, | |
| "params": { | |
| "gt_value": 180, | |
| "tolerance": 5 | |
| } | |
| }, | |
| { | |
| "metric_id": "n3_to_w_interruptions_accuracy", | |
| "type": "numeric_check", | |
| "target_key": "n3_to_w_interruptions", | |
| "weight": 50, | |
| "params": { | |
| "gt_value": 3, | |
| "tolerance": 0 | |
| } | |
| } | |
| ] | |
| } | |
| } | |