Fixed signal distance gave no clear Brier score gain over confidence in eight offline SEEG condition changes

Signal distance gave no clear gain in estimates of reader correctness. The test used 975 later trials from eight public SEEG recordings in OpenNeuro ds004473 version 1.0.2. Each fixed reader separated rest from pre-response windows. We compared correctness models that used confidence alone or confidence with signal distance. The distance used reader fitting data only. The models used a separate early calibration segment. Mean Brier score was 0.210815 for confidence and 0.210553 with distance. Mean gain was 0.000262. The descriptive 95% whole-person interval was -0.000781 to 0.001284. Four people gained and four people lost. The sign-flip reference gave p = 0.695. Distance alone gave mean correctness AUC of 0.509. Fixed decision gates did not establish a benefit. The package includes code, features, three figures, source receipts, and independent checks. All five main result files passed byte replay. The cohort was reused, and trial selection was offline. The result does not test stimulation or all distance methods.

stereo-EEG · brain-computer interfaces · signal distance · calibration · Brier score · selective classification