Pair repair reduced mean Brier error under equal 60 Hz voltage at one training RMS in eight reused SEEG records with simulated contact loss

We tested two gap repair rules in eight reused public SEEG records from OpenNeuro ds004473 version 1.0.2. We retrieved the original contact samples for all 674 later test trials. Each fixed reader used eight bipolar pairs and intact earlier training data. Independent contact packets had a 20 percent drop probability, with boundary packets kept. Actual contact loss was 19.20 percent. We added equal 60 Hz voltage to every contact at one median training bipolar RMS. Mean Brier error was 0.490280 after separate contact repair and 0.268045 after repair with the union mask. The paired difference was 0.222235, with a descriptive 95 percent person bootstrap interval of [0.190610, 0.255434]. All eight records had lower error with pair repair. A shared linear repair operator cancels any equal added reference voltage in exact arithmetic. Pair repair still had higher mean error than the constant probability 0.5 forecast, which has error 0.25 here. The package includes exact signals, masks, probabilities, models, proofs, figures, code, public source receipts, and repeat checks. This was a digital stress test on a reused cohort. It did not test actual data loss, stimulation, or a live BCI.

stereoelectroencephalography · signal processing · brain-computer interface · missing samples · linear interpolation · reference voltage · Brier error