250 and 500 ms SEEG windows gave 75.8% and 75.3% mean balanced accuracy on later trials in eight patients at known keypress times

We compared 250 ms and 500 ms power windows in public SEEG data from eight patients with epilepsy. All response windows ended at known keypress samples. Each model used eight bipolar pairs selected on early trials. Separate later trials gave mean balanced accuracy of 75.8% at 250 ms and 75.3% at 500 ms. The paired mean difference was 0.58 percentage points. The descriptive 95% patient bootstrap interval was -3.67 to 4.58 points. The exact two-sided patient sign-flip p-value was 0.81250. The primary test did not establish a difference. This was not an equivalence test. The analysis used 1,765 selected trials and 671 later test trials in OpenNeuro ds004473 version 1.0.2. The package preserves features for four window lengths, all scores and models, four figures, code, source records, and reproduction instructions. All 1,765 retrieved source ranges passed hash checks. All 500 ms features matched the archived source features exactly. The test measures task-state separation at known event times. It does not measure live detection delay, continuous false alarms, or stimulation effects.

stereo-EEG · brain-computer interfaces · power windows · balanced accuracy · causal filtering