Early phase calibration reduced 50 ms disagreement with an offline theta filter in seven SEEG rest records

Early phase calibration reduced later mean absolute phase disagreement at a simulated 50 ms delay in seven public SEEG rest records. A sinusoid forecast used the last 500 ms of stored samples. Early offline phase targets selected a frequency before the later test. At 50 ms, person mean disagreement was 76.49 degrees for calibration, 84.45 degrees for nominal theta power selection, and 83.33 degrees for fixed 7 Hz. Calibration improved all seven people against power selection and six against fixed 7 Hz. Both planned person sign tests gave Holm p = 0.03125. The power-band boundary changed the result: the calibration gain was only 1.11 degrees against a strict power-bin rule. Also, 65.4 percent of calibrated forecasts differed from the offline target by more than 45 degrees. Constant training windows exposed numerical phase sensitivity in one person. The package includes selected source signals, pinned retrieval instructions, code, results, five figures, and source and repeat checks. This reused sample gives an exploratory offline result. It does not establish biological phase accuracy, stimulation benefit, or live BCI performance.

stereoelectroencephalography · phase forecasting · theta oscillations · frequency calibration · offline validation · neural engineering