A ±6 dB shaft gain bound retained 9.0% of windows after simulated loss in eight SEEG readers with fixed equal contact gains

An output rule kept every accepted label stable under a stated gain bound in eight frozen public SEEG readers. The source was OpenNeuro ds004473 version 1.0.2. The primary bound allowed at most one shaft with gain from -6 to +6 voltage dB. Both contacts had equal positive gain, fixed within each window. Other inputs and reader values stayed fixed. After simulated -6 dB shaft loss, the rule retained 9.0% of windows. The original confidence gate retained 68.7%. These are means across people, with equal shaft weight within each person. The descriptive 95% person bootstrap interval for full retention was 6.8% to 11.2%. The test used 1,950 distinct windows and 35 shaft cases. These gave 8,524 repeated evaluations. The rule had no accepted label flips and no outputs outside the intact reader's accepted-label set. It still accepted 225 wrong task labels among 810 retained cases. Stability does not establish task correctness. Independent checks tested 2.7 million extreme reference scores. Fresh public voltage excerpts checked the gain formula. All 13 main numeric files passed an exact replay. The package includes reusable code, a proof, three figures, frozen inputs, source records, and repeat instructions. The data reuse known parent results. The test uses offline event windows and a narrow gain model. It does not establish live BCI performance or a policy for brain stimulation.

selective classification · confidence gate · signal gain · SEEG · brain-computer interface · robustness