Agreed task labels from separate SEEG shaft groups had 31.0% mean error after condition changes in eight recordings

We tested joint errors of two fixed SEEG task-state readers on separate electrode shaft groups in eight public recordings. Each reader used four bipolar pairs and 32 early fitting trials. The next condition segment supplied 975 test trials. Mean agreement acceptance was 55.2%. Error among accepted windows was 31.0%. Mean joint errors were 16.23%, against 15.74% for a local class-conditioned pairing reference. The mean excess was 0.49 percentage points. Its descriptive 95% person interval was -0.30 to 1.24 points. The exact person sign-flip p value was 0.2578. The paired-trial permutation p value was 0.1733. The primary test did not establish excess errors or error independence. The reference needs exchangeable trial pairs within local blocks. A full reader gave 30.8% mean error at the same accepted count, using a retrospective confidence rank. This exploratory study reused OpenNeuro ds004473 version 1.0.2. Labels required known events, and inherited quality checks used later signal. The package preserves the plan, features, models, predictions, figures, source records, raw retrieval recipes, code, and repeat instructions. All 24 fresh raw ranges matched archived hashes. A separate process reproduced all ten numerical result files.

stereoelectroencephalography · brain-computer interface · joint errors · agreement gate · task-state decoding · error pairing · clinical recordings · exploratory study