A voltage gate reduced 70–110 Hz SEEG repair error but exceeded hold error with simulated endpoint faults equal to voltage RMS in 20 patients

We tested a fixed baseline voltage gate for filter state repair after simulated 13-sample gaps at 4096 Hz. The reused public SEEG data were from 20 patients and 1,671 native bipolar channels in OpenNeuro ds005169 version 1.0.0. The forward filter passed 70 to 110 Hz. The primary fault added one early voltage RMS to the first returned sample. With independent signs by gap and channel, mean patient median error was 118.43% of reference RMS with immediate repair. It was 71.91% with the 99th-percentile gate and 55.45% with hold. The paired decrease from repair was 46.53 points, with a descriptive 95% patient bootstrap interval of 26.37 to 78.41. All 20 patients improved over immediate repair. Seventeen had lower error with hold. With unchanged endpoints, the gate raised error from 25.40% to 26.28%. A simulated common-sign fault and a valid voltage step gave identical first decisions in all 80,208 gap-channel pairs. The package preserves exact inputs, hashes, code, scores, decisions, bootstrap draws, and three figures. All 20 source files passed full hash checks. All 10,096,735 saved voltage values matched the sources. An independent recurrence reproduced the gate outputs. The full analysis repeat matched all saved numerical results. This test concerns recorded waveform error under controlled faults. It does not establish neural origin, live artifact recovery, or decoder benefit.

SEEG · signal processing · filter state repair · voltage gate · simulated endpoint faults · brain-computer interface