Refitted spectrum controls gave a mean 50 ms phase error of 56.10 degrees in seven SEEG rest records with a 100 ms reader and a 2 s offline reference

We tested whether learned theta phase forecasts had less error than spectrum controls. We used 28 selected SEEG pairs from seven people in OpenNeuro ds006065 version 1.0.0. Each control kept block power spectra and cross spectra. We fit the reader again on each of 99 controls per person. The 100 ms reader forecast 50 ms ahead. Mean absolute phase error was 56.59 degrees for real signals and 56.10 degrees for controls. The real-minus-control difference was +0.49 degrees. Its person bootstrap 95% interval was -0.74 to +1.79 degrees. The short-history gain over a 500 ms reader was 1.16 degrees for real signals and 1.26 degrees for controls. An equal-block-duration check gave the same pattern. This test did not show lower error for real signals. The centered 2 s target used future voltage and shared past voltage with the reader. The result does not prove equivalent errors or test stimulation benefit. All seven full source hashes and extracted inputs passed. The package contains figures, voltage, code, models, scores, source checks, and reproduction instructions.

spectrum controls · phase forecasting · SEEG · theta · ridge regression · surrogate data · signal processing