Unreviewed veto calibration raised crossing excess by 8.45 points in 23 SEEG sleep records with a fixed reader

We tested whether a low-frequency voltage veto keeps its benefit when its limit uses calibration without discharge review. We used public SEEG sleep voltage from Falach and colleagues, Figshare 26131978 version 3. We kept the high-gamma reader threshold fixed. In 23 people, 596 adjacent pairs, and 425 later expert labels, full-minute empirical calibration raised event-minus-background crossing excess by 8.45 percentage points relative to reviewed calibration. The 95% interval across people was 4.58 to 12.75 points. Matching calibration sample count did not remove this effect. A fixed median and median absolute deviation rule on log peaks reduced excess by 8.56 points relative to the empirical unreviewed rule. Its mean excess was 4.85 points, compared with 4.97 for reviewed calibration. This does not establish equivalence. The robust rule lost 0.65 points of controlled 90 Hz detections and 3.70 points when a 15 Hz signal was also added. These are test voltages, not intended BCI commands. A secondary test also used unreviewed high-gamma calibration. The source-voltage replay matched all 100 compact arrays exactly. The package contains figures, code, source hashes, metadata, feature arrays, full result tables, controlled trials, and replay checks. The evidence concerns short epilepsy sleep records. It does not show BCI task accuracy or stimulation benefit.

SEEG · intracranial EEG · brain-computer interface · signal processing · calibration · epilepsy · veto threshold · high-gamma power