Local baseline subtraction did not establish higher cued-key accuracy with eight SEEG pairs in later trials from eight people

We tested whether local baseline subtraction improves two-key decoding with eight bipolar SEEG pairs. The source was OpenNeuro ds004473 version 1.0.1. The fixed sample had 1,864 retained trials from eight people. Earlier trials trained each model. The 564 later trials tested it. Mean balanced accuracy was 54.75% with subtraction and 53.27% with absolute power. The paired mean change was 1.48 percentage points. The descriptive 95% person bootstrap interval was -3.18 to 7.25 points. The exact two-sided person sign-flip value was 0.6875, under independence and symmetry assumptions. Four people had higher accuracy with subtraction. These data did not establish a gain. Baseline-only accuracy was 54.47%. A proof shows gain cancellation when each pair has the same positive gain in both windows, above the numerical power floor. This property does not establish better decoding. The package preserves fixed features, all decisions, three figures, code, source versions and hashes, and fresh public raw ranges. All 10,008 audited power values matched after rounding. A full repeat gave 495 byte-identical files. This secondary analysis reuses test trials. It does not validate live BCI control.

SEEG · stereoelectroencephalography · baseline subtraction · balanced accuracy · brain-computer interface · bipolar pairs · ridge regression · signal processing