Fixed theta load templates increased mean session 2 encoding error in 14 SEEG patients under the source reference

We tested whether a template for each letter load predicted encoding signals in another session better than a pooled template. The source was OpenNeuro ds004752, version 1.0.1. The test used the first two numbered sessions from all 14 patients with source hippocampal labels. It included 96 common contacts, 563 training trials, and 579 test trials. Complex coefficients covered 4 to 8 Hz. Test load was known, and the penalty count was fixed at 20. Mean relative error increased by 0.737%. The 95% patient bootstrap interval was 0.468% to 1.025%. Error increased in 13 of 14 patients. The exact two-sided patient sign-flip value was p = 0.000244 under independent patients and symmetric null effects. The pooled template also had more error than a zero-vector prediction in all 14 patients. Power-preserving phase controls gave similar load-template scores. A penalty count of 100 reduced the mean error cost, and its interval included zero. The result applies to these signals and fixed fitting rules. It does not show that the hippocampus has no load signal. The package preserves source versions and hashes, complex features, raw sample windows, code, templates, losses, bootstrap draws, and figures. Independent checks passed. An offline repeat reproduced all 183 model result files. Instructions support an offline repeat and full source retrieval.

stereoelectroencephalography · verbal working memory · theta oscillations · complex Fourier coefficients · cross-session prediction · hippocampus · phase controls