A ridge penalty of one gave no clear beta coherence gain above a phase control after one minute of training in 19 SEEG records

We tested a fixed ridge penalty of one in 19 public SEEG records from OpenNeuro ds003876 version 1.0.2. One minute set the complex channel weights. The next minute tested beta coherence at 13 to 29.5 Hz. We subtracted the exact conditional mean under independent group phase rotations. With equal patient weight, the score gain over unrestricted canonical coherence was 0.003360. The 95% patient bootstrap interval was -0.004486 to 0.016331. Seven patients gained and 12 lost. The two-sided sign-flip test gave p = 0.951271. It assumes independent patients and symmetric null contrasts. This result gave no clear gain. It does not prove equal performance. A secondary 20-second fit showed a gain, but its unrestricted weights were often nonunique. A second BLAS build preserved the primary result and changed some 20-second baseline scores. The package gives all test cells, primary weights, code, and results. Fresh public byte downloads rebuilt all 4,222,800 source coefficients exactly. The package preserves source pins, hashes, retrieval recipes, and reproduction commands. This test uses one selected record per patient and one time split. It does not establish neural interactions or a clinical marker.

ridge regularization · squared coherence · phase control · stereoelectroencephalography · SEEG · intracranial EEG · beta band · canonical coherence · rank · patient bootstrap · OpenNeuro · reproducibility