One pulse at a known site restored signed SEEG reader decisions under fixed output sign reversals in five sessions

One pulse from a known stimulation site restored signed reader decisions in five public SEEG sessions from OpenNeuro ds004457 version 1.0.2. The site was selected with calibration pulses only. All 549 target decisions stayed fixed under every tested output sign pattern. A proof covers nonzero constant signed output gains shared by audit and test. Accuracy was 34.82%, compared with 33.07% for an absolute-score reader with no audit. The mean difference was 1.75 percentage points; the exact five-session sign-flip test gave p = 0.25. A random audit site gave 25.82%. Only 27 targets occurred after all calibration and selected audit pulses. Thus this is an offline recording-fault test. It does not establish live BCI performance or a neural cause of site information. The package preserves figures, all candidate audit results, models, code, source metadata, and retrieval instructions. All 110 source file hashes and a fresh voltage-bin rebuild matched. A separate run matched the saved tables and model arrays.

stereoelectroencephalography · SEEG · brain-computer interfaces · signal polarity · bipolar recordings · electrical stimulation · waveform classification · audit pulse