A temporal voltage correction with alpha = 1 increased mean SEEG forecast error after contact order reversal in eight STReEF patients

We tested whether voltage before a pulse improves a five-pulse SEEG template. We used OpenNeuro ds005448 version 1.0.0, with eight patients and 96 conditions. The later five test pulses had reversed source contact order. The target was centered voltage at 20 to 500 ms, projected onto 16 cosine modes. A shared temporal correction with fixed ridge penalty alpha = 1 increased mean relative test error by 1.95%. The descriptive patient bootstrap interval was 1.11% to 2.81%. Error increased in all eight patients. An early-only penalty choice gave mean skill of -0.047%, with interval -0.239% to +0.136%. Its matched no-pulse control gave +0.603% skill. A stronger fixed penalty gave a small positive result. An added error identity showed that correction size exceeded its alignment benefit at the primary penalty. All 960 raw ranges and 147 metadata files passed source checks. Independent least squares and exact analysis replay passed. The package retains four-window features, fitted maps, all scores, and three figures. GPL analysis and extraction scripts are omitted, so a full local replay and source rebuild are unavailable in the package. This exploratory study reuses a clinical sample. It does not isolate a neural stimulation effect or establish live interface performance.

stereoelectroencephalography · voltage forecasting · ridge regression · brain-computer interfaces · OpenNeuro · STReEF