Assuming independent, exchangeable patients, exact 1 s rotation correction gave an uncertain SOZ AUC gain for guarded 20 ms HFO matches in 18 SEEG records

We made an exact block rotation method for public detector annotations. The method excludes STE events near block edges and computes match distributions on the integer microsecond grid. We used the first sleep hour of 18 SEEG patients from OpenNeuro ds005398 version 1.1.3. The main test used one-second blocks and an inclusive 20 ms match limit. It kept 22,376 of 23,297 STE events. Mean SOZ AUC changed from 0.691225 to 0.697784. The mean gain was 0.006558. Its 95 percent patient bootstrap interval was -0.006171 to 0.019377. This interval assumes independent, exchangeable patients and included zero. The exact sign reversal p value was 0.273438 under independent, symmetric null patient effects. The package includes a proof, source bytes, exact block distributions, code, and reproduction checks. A full repeat reproduced all 21 main output artifacts. These records were used in prior work. This is a method extension, not an independent patient replication. Detector agreement does not establish a neural event source or clinical benefit.

stereoelectroencephalography · high frequency oscillations · seizure onset zone · exact rotation reference · finite-grid distribution · channel ranking · AUC