A common Gamma rate increased mean joint STE HFO forecast loss in 18 sleep SEEG records after an early fit with matched shaft distributions

We tested one common Gamma rate for joint HFO count forecasts across shafts in 18 public sleep SEEG records. OpenNeuro ds005398 version 1.1.3 supplied the released STE marks. We fit on the first 30 minutes and tested the next 30 minutes. Both models had identical single-shaft negative binomial distributions. The common-rate model increased mean loss from 1.687291 to 2.040635 nats per shaft per minute. The increase was 0.353344, or 20.94 percent. Loss increased in 15 records. The patient bootstrap 95 percent interval for the increase was 0.174938 to 0.592424. Population use requires independent, exchangeable patients. All eight fixed checks retained higher mean loss. Two extra sign references did not pass the 0.05 level after Holm correction. A post-result score split assigned 94.72 percent of the net mean increase to count allocation across shafts. This reused clinical cohort limits the scope. The result tests one dependence model and released marks. It does not establish neural independence or clinical benefit. All 94 fresh source inputs matched their hashes. A complete repeat matched all 202 numerical files. The package includes source inputs, proofs, code, counts, fitted parameters, all scores, uncertainty draws, and instructions.

high-frequency oscillations · stereo-electroencephalography · joint count forecasts · Gamma-Poisson model · negative binomial distribution · marginal calibration · forecast loss