An eight-pair SEEG test did not establish lower response detection accuracy on 33 cue–key conflicts than on matched trials in six patients

We tested response detection on 33 cue–key conflicts in six patients from public OpenNeuro SEEG data, version ds004473 1.0.2. Each conflict was matched to one unique cue-matching trial from the same patient, block, recorded key, and task condition. Each eight-pair model learned only from cue-matching trials outside its test block. After the response, mean balanced accuracy was 55.3% on conflicts and 52.6% on controls, with equal patient weights. The difference was +2.64 percentage points; the exact two-sided patient sign-flip value was 0.875. A descriptive patient-and-pair bootstrap interval was −16.95 to +24.44 points. The test did not establish lower or equal accuracy on conflicts. Rest false detection rates were 43.6% and 48.6%, so these results do not establish reliable BCI use. The package preserves all features, source records, matches, 58 models, every score, code, and three figures. Fresh public byte ranges reproduced all conflict and control features exactly. A separate signal reader and ridge solver passed, and a full repeat produced identical core results. This is an offline response-state test, not a test of intent or stimulation.

stereoelectroencephalography · brain-computer interface · response detection · cue–key conflict · matched controls · ridge classification · neural signals