SEEG power before keypresses identified response keys across timing instructions in one of eight patients under conditional tests

We tested response-key transfer between recorded fast and slow instructions in public SEEG dataset ds004473, version 1.0.1. The study used 2,030 trials from eight patients. Each input contained 500 samples before the keypress. A fixed power model excluded complete test time blocks and matched training counts. Only patient 2 passed the corrected conditional label test. Its mean transfer accuracy was 72.99 percent, with corrected p = 0.0009. Cohort transfer and within-instruction accuracies were 53.41 and 50.84 percent. Their mean difference was 2.57 percentage points; exact paired p was 0.13281. The patient bootstrap interval was 0.007 to 5.329 points. These methods gave different decisions. The comparison does not establish a transfer advantage. Tests require exchangeable labels within the defined groups. Cue direction and response key usually matched. Movement could start before the keypress. The findings do not establish motor specificity or online use. All pinned signal ranges were retrieved again. Features and main results matched exactly. The package preserves code, data credits, features, trial labels, predictions, all label draws, independent checks, and reproduction instructions.

stereoelectroencephalography · response-key decoding · instruction transfer · ridge regression · label permutation · time-block cross-validation