Assuming independent patient sampling, the bootstrap interval included zero gain from prior trial theta in later session hippocampal SEEG forecasts in 14 patients

We tested whether prior trial hippocampal theta improved power forecasts in later numbered sessions. We used public OpenNeuro ds004752 version 1.0.1: 14 patients and 62 SEEG sessions. The baseline used current load, prior load, prior error, and trial position. The added model also used prior trial theta. Both models used ridge penalty 10. The first 12 eligible pairs in each test session set the intercept. We scored 1,049 later pairs in 33 sessions. The mean patient gain was −2.146%. Negative gain means greater squared error. The descriptive 95% patient bootstrap interval was [−6.168%, +1.445%]. It included zero and did not establish a mean gain for this fixed learner. The interval assumes independent, exchangeable patients and conditions on the observed fits and sessions. All 12 fixed checks had negative mean gains. Independent fits, fresh raw excerpts from all 62 sessions, and one full raw-session rebuild passed. The package stores data, source versions, hashes, code, forecasts, results, and retrieval instructions. Epoch order does not preserve elapsed trial spacing. This result does not prove absence of neural state dependence.

hippocampus · stereoelectroencephalography · theta power · forecasting · patient bootstrap · working memory