Adding SEEG power did not show a clear wrong-key ranking gain over timing in six recordings with fixed block calibration

We tested whether SEEG power after a response improves wrong-key ranking over task timing. We used OpenNeuro ds004473 version 1.0.2. The fixed sample had 2,012 valid-key trials in eight recordings. Six recordings had 33 wrong-key events. Calibration used cue-key agreement trials from other source task blocks. The model used eight bipolar pairs and three power bands. The primary score added timing and SEEG scores with equal weights. Mean within-block AUC was 0.660 for timing and 0.673 for the combined score. The gain was 1.34 percentage points. A descriptive patient bootstrap interval ran from -1.41 to 4.35 points. The exact patient sign-flip value was 0.500. A control with 200 label draws and full model refits gave 0.358. These tests did not establish a gain. We rebuilt all 33 wrong-key trials and eight agreement trials from pinned public EDF ranges. The rebuilt features matched the inputs exactly. A separate full analysis run reproduced all saved numerical results. The package preserves source features, labels, fitted models, scores, control data, figures, code, and retrieval instructions. The test used physical key responses and offline calibration. It did not test live BCI use or brain stimulation.

stereoelectroencephalography · wrong-key detection · anomaly scoring · brain-computer interface · within-block AUC