Eight SEEG means increased offline forecast loss for later changes of mind in all 14 eligible additional MetaRDK records at penalty 10

We tested a fixed forecast of later recorded changes of mind in 15 additional public MetaRDK SEEG records. Fourteen met the saved class-count rule. Their later test sets contained 1,332 trials and 130 changes. At ridge penalty 10, adding eight selected pre-movement high-gamma means increased test log loss in all 14 records. Equal-person mean loss rose from 0.335932 to 0.387412 nats per trial. The mean increase was 0.051479. Its descriptive 95% person bootstrap interval was 0.032638 to 0.074280. The exact two-sided sign-flip probability was 0.00012207 under independent person effects and a symmetric null. A constant fit-rate forecast had lower mean loss than both models. One fit label had a change time after the final click. Removing that row in a separate post hoc check left higher loss in all 14 records. All 6,500 original probabilities matched exactly. A fresh source rebuild matched one additional record. Independent fits and scores passed. The 23 MB package preserves compact inputs, code, source versions, model states, predictions, three figures, audits, and replay steps. The records come from the same dataset as the earlier six-record study. Source processing and retrospective rules prevent a live BCI claim. The result applies to this fixed offline method and gives no stimulation evidence.

change of mind · stereoelectroencephalography · SEEG · high-gamma activity · logistic regression · forecasting · brain-computer interface · MetaRDK