The Brier score gain from voltage energy before a pulse was uncertain in eight offline SEEG sessions

We tested whether voltage energy before a pulse helped estimate a wrong stimulation-site reading beyond response scores. We used public STReEF SEEG data from eight people with epilepsy. Each person had four selected sites and 20 reversed-order test pulses. Each error model used only calibration error labels from seven other people. No test labels entered any fit. The site reader had 115 correct readings out of 160. Adding before-pulse energy changed mean Brier error from 0.142813 to 0.141587. The gain was 0.001226. The descriptive 95 percent patient bootstrap interval was -0.002015 to 0.004565. This interval conditions on fitted predictions and does not refit shared donor models. At 80 percent coverage, both models kept 100 correct readings out of 128. The added signal did not improve that result. All 320 source epochs and 421,120 features passed a fresh raw-data rebuild. The package contains code, compact data, source records, fitted models, per-pulse results, figures, and rebuild commands. The mixed recording schedule, reused cohort, and possible artifact limit the result. No live feedback or brain-writing benefit was tested.

SEEG · stimulation-site reading · Brier score · error estimation · selective prediction · voltage energy · brain-computer interface