Independent draws from 17 SEEG records gave a 67.16 degree mean error for linear 50 ms offline theta phase forecasts

We tested whether a causal model can predict offline theta phase after independent draws of raw SEEG samples. We used 17 public rest records from seven patients in OpenNeuro ds006065 version 1.0.0. Each record had 20 independent sample controls. The linear model used a past 0.5 second window and earlier fit data. At 50 ms, the equal-patient control mean error was 67.16 degrees. A uniform random phase prediction has an expected error of 90 degrees. The descriptive 95 percent patient bootstrap interval was 67.00 to 67.34 degrees. Real SEEG error was 63.05 degrees. The secondary exact two-sided patient sign-flip test gave p = 0.09375 for real error minus control error. Forecast amplitude ranked control accuracy with AUC 0.612. A finite-kernel proof and Gaussian simulation showed that observed past samples can predict part of this offline target. The future-only target component gave 89.99 degree error in the Gaussian check. Thus accuracy against this target does not require raw sample time order. These results do not measure a clinical effect. The package contains input data, pinned sources, code, models, forecasts, results, proof, and repeat commands. Fresh source byte ranges exactly rebuilt all 17 input arrays. A separate implementation checked all 1,445 fitted models. A second run matched 26,061 numeric arrays, three result tables, and the result summary exactly.

SEEG · theta phase · offline phase · phase forecasting · independent-sample controls · signal processing · neuroscience