A ridge SEEG error score equalled a choice score with binary target sign reversal in 19 offline records

We tested a known binary target in task-error reading from public MetaRDK SEEG. We used 19 records, 2,530 early training trials, and 1,677 later test trials. The primary window was 50 to 450 ms after movement onset. The ridge penalty was 100. Mean target-conditional error AUC was 0.57199 with target multiplication and 0.51625 with the original neural features. The paired gain was 0.05574. Its 95% person bootstrap interval was -0.00349 to 0.11993. The exact sign-flip p value was 0.09716. We proved a score identity under matched training weights, scale, and penalty. The target-based error score equals a choice score multiplied by the negative target sign. This identity held in all 95 fitted conditions. Conditional error and choice AUCs were also equal. Thus, error AUC alone cannot identify a distinct neural error signal. The primary gain was uncertain. We freshly rebuilt one full pinned public signal record and checked all 44 source metadata files. Independent calculations checked 285 error fits and every AUC. A full numerical replay passed. The package preserves compact data, models, predictions, figures, source receipts, and reproduction code. The same clinical cohort supports earlier studies. Source processing and trial selection are offline. No live BCI or stimulation benefit was tested.

stereoelectroencephalography · brain-computer interface · error decoding · binary target · ridge regression · area under the curve · neural signals