Later model fits gave no clear gain in 50 ms offline theta phase forecasts in SEEG rest from seven people

We tested whether an early model fit in a later rest file improved 50 ms theta phase forecasts. We used public SEEG data from seven people in OpenNeuro ds006065 version 1.0.0. Four voltage pairs per person stayed fixed from the earlier file. All methods used the same later test times. The offline target used a centered 2-second theta filter. Mean absolute error was 57.26 degrees for the frozen earlier model and 57.77 degrees for the later fit. The mean gain was -0.51 degrees. Its 95% person bootstrap interval was -1.43 to +0.25 degrees. The exact two-sided sign-flip p value was 0.328. A later phase-offset update also gave no clear gain. About half of forecasts were within 45 degrees of the target. These results do not establish a calibration benefit or model equivalence. The package preserves voltage excerpts, models, forecasts, source hashes, code, figures, and repeat checks. It provides a test for calibration transfer. It does not test delivered pulses or a causal stimulation effect.

theta phase · SEEG · intracranial EEG · phase forecasting · calibration transfer · brain-computer interface · ridge regression