Half-strength prediction increased mean relative squared error against zero prediction for beta triangle cycles in paired 60-second SEEG files from 36 epilepsy patients

We tested a phase-invariant cycle measure in public OpenNeuro ds004100 version 1.1.3. Two interictal files from each of 36 epilepsy patients supplied paired 60-second SEEG samples. We chose one fixed bipolar pair per named shaft. The 453 signals gave 11,326 triangles. The primary score was the cosine between beta cycle vectors. Its patient mean was 0.04867. The 95 percent patient bootstrap interval was 0.00088 to 0.11777. The label reference tail was 0.018 under uniform whole-node permutations. A fixed secondary predictor used half of each file 1 score to predict file 2. Mean relative squared error increased by 22.35 percent against zero prediction. Its 95 percent interval was 16.97 to 27.63 percent. Error increased in 34 patients. Phase sine and distinct-window checks had intervals that included zero. A post-result diagnostic removed the largest positive contributor, who had one triangle. The remaining mean cosine was 0.02149, with an interval of -0.00553 to 0.05256. The bootstrap assumes independent, exchangeable patients. The result does not support the tested transfer rule or prove a neural circuit. It does not exclude other predictors. The package preserves all Fourier data, source locks, code, proofs, scores, figures, and reproduction instructions. All 144 raw ranges passed fresh hash checks. Independent calculations passed. An offline repeat matched all 39 fixed numeric files exactly.

stereoelectroencephalography · SEEG · phase coupling · coherency · triangle cycles · repeatability · epilepsy