One gain rule per shaft gave lower mean later Brier error than one rule per pair in eight SEEG records under an ideal equal contact gain model

We tested two gain rules for a brain signal reader. We used public OpenNeuro ds004473 version 1.0.2. Eight reused SEEG records supplied 1332 training trials and 674 later test trials. Each input had eight bipolar pairs and three band powers. The shaft rule kept 18 to 21 power coordinates. The pair rule kept 16. Both rules cancelled ideal positive contact gains that were equal on each shaft and constant within each buffer. Mean later Brier error was 0.187576 for the shaft rule and 0.190209 for the pair rule. The paired change was -0.002633. Seven of eight people had lower error. The descriptive 95% person interval was -0.006120 to -0.000266. The exact sign-flip reference was 0.07031. The fixed-penalty interval included zero. Full power had lower mean error without added gain. The package includes a gain proof, code, source records, all fits and test scores, and three figures. All 24 numerical files passed an exact repeat. Fresh source ranges matched 32 input windows. This is an offline method test on a reused clinical sample. It does not establish a live BCI benefit or a neural cause.

stereoelectroencephalography · signal processing · gain invariance · Brier score · brain-computer interface · bipolar electrodes