Mean inferred choice AUC was 0.457 with time changes and 0.511 with window means in 9 offline smoothed SEEG tests before movement

We tested whether time changes inside a 400 ms SEEG window improve later choice scores at the same eight pairs. We used public OpenNeuro ds006253 version 1.0.3. The signals had source smoothing. The sample has 9 records from an earlier cohort, with 1148 early fit trials and 761 later test trials. We compared eight window means with the same means plus three changes per pair. We selected pairs and tuned each reader with early data only. Before mouse movement, the mean AUC was 0.511 for means and 0.457 for means with changes. The paired difference was -0.053. Its 95% person bootstrap interval was [-0.086, -0.025]. The exact two-sided sign-flip reference gave p=0.0078. This interval does not establish equal reader performance. We preserve four-part features and complete source hashes. We also preserve reader models, trial predictions, code, and figures. Choice was inferred from stimulus direction and correctness. The test cannot separate choice from task error within one stimulus direction. Event alignment and source smoothing do not establish online input access. The test gives no evidence about brain stimulation.

stereoelectroencephalography · choice decoding · time-window features · logistic regression · offline brain-computer interface