Maps from 16 trials in a new task condition gave 0.028 bits more mean loss than matched earlier maps in eight fixed SEEG decoders

We tested whether 16 trials after a task condition change improve the confidence values of fixed SEEG decoders. Public OpenNeuro ds004473 version 1.0.2 supplied one recording from each of eight people with epilepsy. We compared a positive confidence scale from the new condition with a scale from 16 earlier trials per person. The shared later test had 711 trials and equal numbers of response and rest windows. Mean loss was 0.894 versus 0.867 bits per window. The mean change was +0.028 bits. The 95% person bootstrap interval was -0.006 to +0.062 bits. Seven people had higher loss with the new scale. Mean balanced accuracy stayed at 64.1% because all class decisions stayed fixed. These reused, selected windows did not show a mean benefit from new labels. The result does not establish equal loss or a harmful effect across people. The package preserves features, source versions, maps, predictions, figures, and code. Fresh source checks, independent calculations, and an exact repeat passed. The test concerns offline task state. It does not measure continuous intent, clinical benefit, or brain stimulation.

stereoelectroencephalography · brain-computer interface · probability calibration · confidence · task condition · log loss · epilepsy