Four selected SEEG channels gave higher mean recall forecast error than regional power in 13 patients under source trial order

We tested a fixed four-channel SEEG rule for early forecasts of later recall errors. We used OpenNeuro ds006136 version 1.0.1, with 14 sessions from 13 patients. We assumed that source trial IDs give task order. Only training trials set channel ranks and model parameters. On 751 later test trials, patient mean Brier error was 0.187537 for four selected channels and 0.179663 for regional power. The mean gain was -0.007874, with a descriptive 95 percent patient-bootstrap interval of [-0.017901, 0.002269]. Three of 13 patients improved. Mean error also exceeded task context and the random four-channel reference. The intervals include zero. This is an exploratory comparison on reused test trials. Full-trial source processing limits live use. The package preserves code, pinned source hashes, matched features, all forecasts, 1,400 random-set models, and three figures. A complete source rebuild and a repeated analysis matched the numerical files exactly.

stereoelectroencephalography · working memory · recall forecast · Brier error · channel selection · brain-computer interface