Coefficients fitted to earlier baselines raised later beta log power variance within task groups in eight ds004473 SEEG patients

We tested whether coefficients fitted to earlier trial baselines reduced later response power variance. The source was public SEEG data from eight epilepsy patients in ds004473 version 1.0.1. The fixed forward test used 986 training trials, 1,017 test trials, and 831 bipolar pairs. The primary measure was beta log power variance within response-side and instruction groups. Fitting reduced training variance by 1.032 percent. It increased later variance by 0.698 percent compared with no correction. All eight patients had an increase. The descriptive 95 percent patient bootstrap interval for the increase was 0.439 to 0.970 percent. The exact two-sided sign-flip probability was 0.0078125. This test requires independent patients and symmetric gains under the null. The result does not establish signal accuracy or clinical benefit. We rebuilt all 4,060 feature epochs from pinned public source ranges. All 839,040 power values matched. The package preserves source code, features, task metadata, source hashes, statistical draws, independent checks, and reproduction instructions. The same patient cohort was used in the cited earlier work.

stereoelectroencephalography · baseline correction · spectral power · beta band · forward validation · variance · trial-level analysis