Datasets:
File size: 6,504 Bytes
d846fa6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | {
"meta_info": {
"case_id": "Mumtaz2016_01.edf",
"bench_subset": "NeuroBench-Core",
"difficult": 1.5,
"original_dataset": "Mumtaz2016"
},
"agent_input": {
"data_path": "data/core/Mumtaz2016_01.edf",
"instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal."
},
"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\nReturn a placeholder JSON object for pipeline compatibility.\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 or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}",
"metrics": [
{
"metric_id": "invalid_pipeline_logic_consistency",
"type": "semantic_consistency",
"target_key": "full_text",
"weight": 100,
"params": {
"judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language."
}
}
]
}
}
|