An early theta phase model reduced later log gamma error in two of seven SEEG rest records

We tested whether early theta phase terms reduce later log gamma amplitude error beyond theta amplitude. We used 28 fixed bipolar pairs from seven public SEEG rest records in OpenNeuro ds006065 v1.0.0. Each pair used 2 to 58 seconds for fitting and 122 to 238 seconds for testing. The mean relative error decrease was 0.134%. The descriptive 95% person bootstrap interval was -0.086% to 0.492%. Two of seven person means and seven of 28 pair means were positive. The exact two-sided sign-flip value was 0.8125. The observed mean exceeded all 999 shifted phase means under a shift-invariance assumption. This control tests phase alignment, not a consistent gain against the amplitude baseline. The filters use future voltage. The result does not establish neural coupling, an optimal stimulation phase, or a live interface benefit. The package preserves exact source-verified traces, features, code, coefficients, predictions, figures, and complete numeric replay checks.

theta phase · gamma amplitude · stereoelectroencephalography · SEEG · phase-amplitude coupling · transfer learning · calibration · brain-computer interface