Mean hippocampal theta templates increased average relative error during encoding and late maintenance between two sessions in 12 SEEG patients

We tested a load-specific mean theta template on public SEEG data from OpenNeuro ds004752 1.0.1. The sample has 12 epilepsy patients, 44 matched hippocampal contact pairs, and 1,085 eligible trials. Session 1 supplied the template. Session 2 supplied all primary test trials. The reference predicts zero complex coefficients at 4 through 7 Hz. Mean prediction gain was -4.988 percent during first encoding and -6.380 percent during late maintenance. The encoding-minus-maintenance contrast was 1.392 percentage points. Its pointwise 95 percent t interval was [-3.619, 6.403]. The exact two-sided sign-flip probability was 0.635254. This test assumes independent patients and sign symmetry under the null. The t interval assumes approximately normal patient contrasts. The result gives no clear evidence of an encoding advantage. It does not establish equal gains or absent event responses. Finite training samples and session change can make template prediction worse. One digital-bound sensitivity check was unstable. The package contains four-phase coefficients, raw examples, templates, trial errors, patient values, code, source versions, range hashes, and repeat instructions. Independent calculations checked all 4,668 selected raw windows and the primary result. A fresh process reproduced all 134 numerical result files byte for byte. This is an extension in the parent sample, not a new cohort.

hippocampus · theta oscillations · stereoelectroencephalography · working memory · held-out prediction · complex Fourier coefficients · template prediction · session transfer