Channel delays fitted in the first 30 minutes increased mean 5 ms detector agreement in the next 30 minutes over patient delays in 18 SEEG sleep records

We tested whether earlier channel delays improve later agreement between released STE and MNI high frequency oscillation starts. We used all 18 SEEG sleep records in OpenNeuro ds005398 version 1.1.3, with 1,285 good channels. We fitted median MNI minus STE delays from 20 ms pairs in the first 30 minutes. The channel rule required five earlier pairs and used the patient median otherwise. It had channel fits on 184 channels. In the next 30 minutes, the mean fraction of STE starts matched within 5 ms increased from 20.09% with patient delays to 21.21% with channel delays. The paired gain was 1.11 percentage points. The 95% patient bootstrap interval was 0.32 to 1.93 points; population use requires independent, exchangeable patients. Twelve records improved, four declined, and two did not change. The primary gain over shaft delays and gains at 2 and 10 ms had intervals that included zero. Alternative agreement scores, exact delay assignment controls, and a chance-time control are preserved. Most marks remained unmatched. This is a test of released timestamp agreement, not neural event validity or clinical accuracy. Earlier full-hour work used this same cohort and informed the question, so the later interval was held out from this fit but was not a blind validation set. All 94 fresh public input objects matched their pinned versions and hashes. A fresh-source run reproduced all numerical files exactly. The package includes inputs, source versions, code, fit and test pairs, patient results, proof, controls, figures, and independent checks.

high frequency oscillations · stereoelectroencephalography · detector agreement · event timing · channel delays · intracranial EEG