Confidence groups decreased the mean 90% target error arc score in one of two SEEG rest recordings for 50 ms theta phase forecasts with an offline reference

We tested phase error limits with public rest SEEG from two people in OpenNeuro ds005670 version 1.0.0. We used eight pairs chosen from early voltage. The forecast target was a fixed offline theta filter, 50 ms after each cutoff. Separate time blocks fit the forecast, confidence groups, and error limits. Four confidence groups decreased the mean 90% target error arc score from 321.66 to 312.86 degrees, a 2.74% decrease. One person had a lower score. The other had a slightly higher score. Group coverage was 89.66%, and mean arc radius was 136.14 degrees. No 90% target group radius was 45 degrees or less. Using one calibration error per second reduced the mean gain to 0.53%. These descriptive results apply to the two recordings and the selected pairs. They do not establish a population effect, a coverage guarantee, or precise live stimulation timing. The package includes selected native voltage, fixed models, a forecast tool, source hashes, tables, and three figures. Both full source files passed hash and timing checks. A complete analysis repeat gave exact numeric equality.

SEEG · theta phase forecasting · confidence calibration · error limits · brain-computer interface