A fixed pre-pulse adjustment gave 70.8% accuracy versus 71.4% without adjustment for neighboring sites at 20–100 ms in five offline SEEG sessions

We tested a fixed pre-pulse adjustment on public SEEG data from five patients with epilepsy. The offline task contained 122 two-site problems. Neighboring sites shared one stimulation contact. Each site supplied four calibration pulses and four separate test pulses. With a 20 to 100 ms response, equal-patient mean accuracy was 70.82% with adjustment and 71.44% without it. The change was -0.62 percentage points. Its descriptive patient bootstrap interval was -1.45 to +0.21 points. An exact patient sign-flip calculation gave p = 0.3125 under independent, sign-symmetric patient contrasts. The later window gave a small gain with large uncertainty. Wrong-pulse pre voltage gave a similar main score. Calibration error fell, but test voltage error rose in four patients in an exploratory check. All 110 pinned source files passed hash checks. Fresh raw extraction and analysis replay matched the saved data and results. The package includes code, source bins, full predictions, three figures, and reproduction instructions. The site schedule confounds this split. The result does not establish a causal brain-state effect or live BCI performance.

stereoelectroencephalography · brain-computer interface · site identification · pre-pulse voltage · conditional response model