A causal linear readout reduced 20 ms disagreement with an offline theta phase reference against a calibrated harmonic fit in seven SEEG rest records

We tested whether an early learned linear readout improved theta phase agreement on later public SEEG data. The source was OpenNeuro ds006065 version 1.0.0. Seven epilepsy records supplied one anatomical proxy pair each. The readout used 0.5 seconds of past raw bipolar data to forecast 20 ms ahead. All settings came from early data. The fixed later test used 190 to less than 290 seconds per record. Against a two-second offline complex reference, mean absolute disagreement was 52.31 degrees for the learned readout and 69.63 degrees for a calibrated harmonic fit. All seven person contrasts favored the readout. The mean reduction was 17.31 degrees. Its descriptive person bootstrap interval was 14.09 to 20.75 degrees. An early magnitude gate accepted 51.05 percent of later times and gave 39.87 degrees of disagreement. Large errors remained. The package preserves source contacts, pinned retrieval recipes, all forecasts, coefficients, code, four figures, and strict repeat checks. This is a new later-file test in a previously studied cohort. It establishes reference agreement against the stated fixed-frequency comparator. It does not establish true brain phase, live timing, stimulation effects, or superiority to stronger causal methods.

stereoelectroencephalography · theta phase · causal prediction · phase estimation · ridge regression · brain-computer interface · offline reference · epilepsy · signal processing