{
    "meta_info": {
        "case_id": "BCIC2020-3_01.edf",
        "bench_subset": "NeuroBench-Core",
        "difficult": 2.0,
        "original_dataset": "BCIC2020-3"
    },
    "agent_input": {
        "data_path": "data/core/BCIC2020-3_01.edf",
        "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR."
    },
    "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 two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": <float|null>, \"abr_window_2\": <float|null>}",
        "metrics": [
            {
                "metric_id": "abr_window_1_accuracy",
                "type": "numeric_check",
                "target_key": "abr_window_1",
                "weight": 30,
                "params": {
                    "gt_value": 0.431956493319918,
                    "tolerance": 0.0215978246659959
                }
            },
            {
                "metric_id": "abr_window_2_accuracy",
                "type": "numeric_check",
                "target_key": "abr_window_2",
                "weight": 30,
                "params": {
                    "gt_value": 0.4287180440931179,
                    "tolerance": 0.021435902204655896
                }
            },
            {
                "metric_id": "attention_state_consistency",
                "type": "semantic_consistency",
                "target_key": "full_text",
                "weight": 40,
                "params": {
                    "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False."
                }
            }
        ]
    }
}