A fixed high gamma SEEG key model gave 48.9 percent mean balanced accuracy with random splits and 48.1 percent on later trials in eight patients

We tested whether random trial splits gave higher response key accuracy than tests on later trials. SEEG means stereoelectroencephalography. We reused 1,988 public trials from eight patients in OpenNeuro ds004473 version 1.0.2. The fixed ridge model used trial order and high gamma power changes at 822 contact pairs. Mean balanced accuracy was 48.90 percent over random splits and 48.13 percent on later trials. There were 100 random splits per patient and condition. The mean difference was 0.77 percentage points. Its descriptive 95 percent patient bootstrap interval was -1.73 to +3.04 points. The exact two-sided sign reversal test gave p = 0.578 under patient independence and symmetric null contrasts. The test did not detect a clear split difference. It does not prove equal performance. The weak key prediction limits conclusions about stronger decoders. This is the same cohort as the earlier feature study. The package preserves powers, labels, every split and test score, primary model fits, code, raw signal fixtures, source hashes, and reproduction instructions. Independent solver and Fourier checks passed. A full repeat analysis matched the results.

stereoelectroencephalography · SEEG · high gamma · response key decoding · ridge regression · random splits · temporal validation · balanced accuracy