A 5% rest error target gave 27.7% response sensitivity on later trials in eight SEEG recordings

We tested a SEEG response decoder at low error targets in eight patients with epilepsy. Public OpenNeuro ds004473 version 1.0.2 supplied 2,032 retained trials. Separate early trials fitted the model and set thresholds. The last 816 trials tested them. With a 5% rest error target, equal-patient means were 3.0% rest errors, 6.9% waiting errors, and 27.7% response sensitivity. The primary waiting-minus-rest contrast was 3.96 percentage points; the exact two-sided sign-flip test gave p = 0.141. This did not establish a common state effect. Calibration on rest and waiting reduced trials with either negative call from 9.7% to 3.2%, and reduced response sensitivity to 14.2%. The package provides a reproducible threshold benchmark, three figures, all features, scores, weights, code, and source versions. Independent model checks, 48 raw sample checks, and full feature retrieval for one patient passed. This is an offline test within recordings. It does not measure continuous interface use or stimulation effects.

stereoelectroencephalography · brain-computer interfaces · threshold calibration · response sensitivity · ridge regression · conformal prediction