Full correction of the largest 20% of calibration change directions increased mean later SEEG covariance error over uniform scaling in 36 patients with 25 s blocks at 1–80 Hz

We tested a covariance correction in two public interictal SEEG records from each of 36 patients. SEEG means stereoelectroencephalography. Calibration used a 25 s block from each record. A separate later 25 s block from each record gave the test score. The correction used the largest 20% of observed change directions at full strength. The primary band was 1 to 80 Hz. The regularization fraction was 0.01. Mean patient squared error increased by 23.98% over uniform covariance scaling. The 95% patient bootstrap interval for this increase was 14.13% to 33.92%. Error fell in 9 of 36 patients. All nine fixed band and regularization checks had a negative mean reduction. A half-strength correction had a small descriptive gain. The target calibration block alone had a larger descriptive gain. No test score selected the primary model or a reported setting. Fresh checks rebuilt all 72 covariance files from 866,839,040 public raw bytes. All 5,184 scores passed separate numerical and contrast-basis checks. The offline repeat matched all numeric outputs. The package preserves code, data, metadata, source locks, results, and instructions. The evidence applies to short excerpts from this selected epilepsy cohort. It does not establish a cause for signal change or clinical benefit.

stereo-electroencephalography · covariance correction · signal processing · covariance distance · calibration · patient bootstrap