Recent calibration did not show a clear accuracy gain in eight SEEG recordings with 32 matched calibration trials

We tested whether recent calibration improved later response versus rest classification in eight public SEEG recordings. Early and recent samples each had 32 trials with the same fast and slow task counts. Mean balanced accuracy was 62.90% for early samples and 65.75% for recent samples. The mean change was +2.86 percentage points. A descriptive 95% patient interval was -0.38 to +6.38 points. The exact two-sided patient sign-flip test gave p = 0.171875 under independence and sign-symmetry assumptions. This test did not establish an accuracy gain or equal accuracy. Three planned secondary checks also did not establish a gain. The package supplies a matched calibration test, all sample lists, 872 model states, scores, four figures, and reproduction code. It preserves complete feature data, pinned source records, raw audit ranges, and fresh retrieval checks. These are retrospective tests with known response times and after-response quality rules. They do not establish useful online control or a stimulation benefit.

stereoelectroencephalography · brain-computer interface · calibration · balanced accuracy · response detection · epilepsy · machine learning