Second harmonic phase terms increased later Brier loss for 50 ms theta confidence against offline targets in 17 SEEG rest files from seven patients

We tested confidence for 50 ms theta forecasts in 17 public hippocampal SEEG rest files from seven patients. The outcome was error at most 45 degrees against an offline 4 to 10 Hz reference. Each file used earlier data to fit probabilities and separate later data to test them. The baseline used forecast amplitude. Added terms used the second harmonic of forecast phase and its interaction with amplitude. Patient mean Brier loss increased from 0.236642 to 0.237731. Mean gain was -0.001089. Six patients had higher loss. The descriptive 95% patient bootstrap interval for gain was -0.001981 to -0.000406. The two-sided patient sign-flip p was 0.03125 under the stated null assumptions. All fixed sensitivity checks had higher mean loss. A marked post-plan calculation found lower calibration loss in every file. The test does not exclude other phase models or test a stimulation effect. The package preserves native inputs, source pins, code, fitted models, query predictions, inference draws, figures, and instructions. New downloads reproduced all 17 inputs. An offline repeat reproduced all analysis arrays and tables exactly.

phase prediction · Brier loss · confidence calibration · theta rhythm · SEEG · signal processing