The mean recall prediction gain from delay high gamma was uncertain on later trials in seven patients with recordings marked as SEEG

We tested whether delay high gamma power adds recall information beyond fixation power and trial order. We used OpenNeuro ds006136 version 1.0.1. The primary sample had seven Utah patients and 473 later test trials. Each channel model was fit on earlier trials. The mean log-loss gain was 0.009554 nats per trial. The 95% patient interval was [-0.004385, 0.026549]. The exact two-sided sign-flip p value was 0.328125. The evidence does not establish a population gain. The files declare SEEG, but the source paper has conflicting implant descriptions. The package preserves 19,617 channel-trial features, exact source hashes, source metadata, models, predictions, and reproduction code. A trial join audit found that 87.60% of compared channel-row positions had different trial IDs. A separate run from retrieved signals gave identical feature and result files. This is an offline test within patients, conditional on source preprocessing.

working memory · recall prediction · high gamma power · intracranial EEG · logistic regression · log loss · patient bootstrap · OpenNeuro