An exact sign-flip test passed at 70 of 86 SEEG sites with six retained pulses and a 20–500 ms window in eight people from ds004696

We tested a small-pulse response screen for brain-computer interface calibration. We used public SEEG data from OpenNeuro ds004696 version 1.0.1. Eight people supplied 86 stimulation sites and 886 candidate recording-pair instances. An exact whole-pulse sign-flip test checked all candidate pairs at each site together. With six retained fit pulses, 70 sites passed in the 20–500 ms window. Their mean after-versus-before AUC on four later pulses was 0.978, with equal weight for each person. Five fit pulses could not pass at a 0.05 threshold because the smallest exact p-value was 0.0625. The six-pulse limit was 0.03125. A separate before-pulse screen passed at 5 sites. The 100–500 ms check passed at 59 after-pulse sites. The within-site error bound assumes joint sign symmetry for null pulse vectors. This assumption was not established by the data. Repeatable stimulation artifact can pass. The study does not show causal neural responses or clinical benefit. The package preserves exact inputs, code, individual scores, four figures, a proof, source pins, and repeat instructions. Separate full-sign loops and direct waveform checks passed. One full source file was freshly retrieved and checked.

SEEG · electrical stimulation · exact sign-flip test · response screening · brain-computer interface calibration · AUC · multiple testing · artifact · OpenNeuro