A 95 percent bootstrap interval included zero gain from joint shaft sampling for one-minute HFO total forecasts in 18 SEEG records, assuming independent, exchangeable patients

We tested forecasts of released HFO total counts in 18 public SEEG sleep records from OpenNeuro ds005398 version 1.1.3. The sample had 1,285 eligible bipolar channels on 200 named shafts. The first 30 minutes supplied the forecasts. The next 30 minutes supplied 540 one-minute test totals. Joint sampling retained measured dependence between shafts. Independent sampling removed that dependence. Both methods had the same shaft count laws and the same expected total. Mean scaled count score was 0.314675 for independent sampling and 0.314564 for joint sampling. Lower is better. Mean gain was 0.000111. The 95 percent patient bootstrap interval was -0.010607 to 0.007580. It included zero under an independent, exchangeable patient model. Joint forecast variance was larger in every record. The median variance ratio was 2.012. Interval coverage rose from 62.41 to 80.74 percent. Both values remained below their forecast interval mass. All eight fixed checks are reported. The result concerns detector marks and this reused cohort. It does not establish a neural cause or clinical benefit. The package contains source bytes, code, exact finite forecast laws, scores, checks, and repeat instructions.

high-frequency oscillations · stereoelectroencephalography · count forecasting · dependence · continuous ranked probability score · patient bootstrap