Added beta power gave minus 0.038 percent mean next-trial error reduction in eight SEEG patients under fixed ridge models

We tested whether beta power after a button response improves prediction of the next trial's reaction time. We used public OpenNeuro ds004473 version 1.0.2 and eight patients with clinical SEEG implants. We reused 1,875 retained trials and formed 1,720 consecutive observed transitions. Earlier trials trained separate patient and condition ridge models at penalty 10. The primary test used 604 later transitions from 13 valid fits. Training selected the 10 percent of pairs with the largest mean response beta decrease. The base model used context, previous reaction time, and previous baseline and response beta levels. The full model added beta power centered about 0.8 seconds after the prior response. The target was natural log next reaction time in seconds. The mean patient error reduction was -0.038 percent. The patient bootstrap 95 percent interval was -1.040 to 0.955 percent. The exact two-sided sign-flip p value was 0.859375. Five of eight patients had positive gains. The interval assumes independent patient sampling. The test also assumes symmetric null gains. The primary test did not show a clear added gain under the fixed models. It does not prove equal models or absent beta function. The next cue was 4.812 to 9.216 seconds after the prior response. This study does not test a second movement during the same high-beta state. Independent regression and leakage checks passed. A second analysis matched all nine numerical output files. Fresh pinned EDF spans checked 31,236 power values in eight test-source trials. All 89 saved metadata files matched the fixed public source commit. The package preserves code, source records, derived powers, transition features, predictions, results, and reproduction instructions.

beta power · reaction time · stereoelectroencephalography · ridge regression · cross-trial prediction · motor behavior