HFO overlap onset order within 10 ms had 50.2 percent mean prediction accuracy across sleep records from seven SEEG patients

We tested whether other sleep records predict onset order for fast ripple marks that overlap ripple marks. We used public SEEG data from OpenNeuro ds003498 version 1.1.1. Seven patients supplied 135 records of exactly 300 seconds. We tested all channel pairs on separate shafts with no shared contact. The primary onset difference was greater than zero and at most 10 ms. Each target direction was trained on other records from the same patient. Training ties received half credit. There were 1,729 ordered event pairs. Mean patient accuracy was 50.2 percent. The descriptive patient bootstrap 95 percent interval was 47.0 to 53.4 percent. A null model independently reversed each whole record and refitted all predictions. It kept every close event count fixed. Of 9,999 draws, 4,737 reached the observed mean. The one-sided p value was 0.4738 under independent uniform record orientations. The primary test did not support repeatable direction under that model. It did not establish absent direction. Mean prediction coverage was 71.7 percent. A fixed 25 ms check gave 55.6 percent accuracy; this was descriptive. The package preserves the method, local plan, source files, hashes, code, full counts, predictions, draws, and reproduction instructions. All 26 numerical files matched an offline rerun. Fresh public retrieval verified all 185 event files.

high frequency oscillations · stereoelectroencephalography · event order · onset timing · held-out prediction · whole-record reversal · epilepsy