Moving a 500 ms SEEG window from a 50 ms to a 300 ms response lead raised reader error in eight recordings

This study tested how input time changes a fixed SEEG reader. It used 2,014 trials from eight public recordings in OpenNeuro ds004473, version 1.0.2. Each input window had 500 samples. Moving its end from 50 ms to 300 ms before the source Response event raised mean Brier error from 0.190 to 0.240. Error rose in all eight people. The paired mean change was 0.050, with a descriptive 95% person bootstrap interval of 0.028 to 0.074. Mean balanced accuracy fell from 70.7% to 63.8%. Training at 300 ms gave 62.0% mean accuracy and did not show recovery. The package adds a six-by-six time transfer benchmark, a cue audit, exact features, model weights, test probabilities, figures, and source retrieval code. All 2,014 fresh source ranges passed their hash checks. A clean numerical replay matched the saved results. The cohort and trial split come from prior sci work. This is an exploratory extension. Known Response events define the windows. Earlier inputs contain less time after the cue. The result does not establish early movement intent or continuous BCI performance.

stereoelectroencephalography · brain-computer interfaces · time transfer · Brier error · balanced accuracy · response lead · cue audit