Two voltage checks accepted fewer correct responses than one check in five offline SEEG sessions at a 5% background calibration target

We tested a voltage check for feedback after brain stimulation. We used public OpenNeuro ds004457 version 1.0.2 SEEG data from five people with epilepsy. The main sample had 134 good stimulation sites on shafts separate from the recording shafts. Each site had three fit pulses, one calibration pulse, and four later test pulses. Both rules used a 5% background calibration target. A site check plus a response check accepted 23.1% of correct responses. A stricter site check alone accepted 28.1%. The mean paired change was -5.05 percentage points. Its descriptive patient bootstrap interval was -9.22 to -1.36 points. Matched before-pulse acceptance was 1.1% and 5.7%, respectively. One patient had too few calibration windows for a finite threshold, so both rules rejected all inputs. Only 32 of 536 test pulses followed all calibration data. These results show a rejection tradeoff in this fixed offline sample. They do not establish a live BCI benefit or a 5% error guarantee. The package preserves voltage inputs, all models and decisions, three figures, source hashes, code, and repeat instructions. Fresh raw recovery, independent arithmetic, and exact numerical replay passed.

SEEG · brain-computer interfaces · electrical stimulation · response detection · voltage signals · background calibration · rank thresholds