Adding delay theta imaginary coherence increased mean recall prediction error in fixed block models from 12 SEEG patients

We tested whether delay theta coupling adds recall prediction value beyond local power and trial position. The source was public OpenNeuro ds006136 version 1.0.1. The source sidecars label the signals as SEEG. The fixed block test used 12 patients with epilepsy, 376 signal pairs, and 1,839 distinct test trials. The added feature was mean squared imaginary coherency at 4 through 8 Hz. Mean patient relative error reduction was -0.5152 percent. The 95 percent patient bootstrap interval was -1.4337 through +0.1936 percent, under independent, exchangeable patients. The exact two-sided sign flip probability was 0.3501, under independent patients and null sign symmetry. Six patients had lower error. Both main models had higher mean error than a training-mean reference. The test gave no clear prediction benefit. It did not establish absence of theta recall information. The package preserves joined features, Fourier inputs, all fits and predictions, source hashes, retrieval code, and repeat instructions. Independent calculations checked all coupling rows and primary model folds. Full extraction and analysis repeats matched. The result applies to these processed trials and fixed models. It does not establish a causal memory mechanism.

working memory · SEEG · imaginary coherency · theta coupling · recall prediction