Early cross-power correction reduced mean later bipolar SEEG curvature error against early parent spectra under source average reference: 24 participants, 20–100 Hz, excluding 55–65 Hz

We tested whether first-minute cross-power helped to predict last-minute bipolar SEEG curvature. We used public OpenNeuro ds005624 version 1.0.0 and the source common average reference. The main band was 20–100 Hz, with 55–65 Hz removed. The test used 24 participants, 34 valid records, and 2,303 disjoint pair-records. Each participant had equal group weight. Adding the early relative real cross-power term reduced mean squared curvature error by 69.4 percent against the mean early parent-contact spectrum. The 95 percent participant bootstrap interval was 51.4–81.8 percent. Error decreased in 23 of 24 participants. The own correction also reduced error by 57.0 percent against a correction from other pairs on the same shaft. The corrected estimate equals the pair's own early bipolar curvature. The result concerns temporal stability of a reference-dependent shape component. It does not identify a neural cause. The interval assumes independent, exchangeable participants. This is a new analysis of the source cohort, not an independent cohort replication. The package preserves code, input spectra, raw examples, source range hashes, results, a local plan, and reproduction instructions.

stereo-electroencephalography · cross-power · spectral curvature · common average reference · temporal prediction