Frequency weights raised mean minimum site coverage from 63.5 to 89.8 percent in five SEEG sessions when 80 percent of simulated commands used one shaft and site responses stayed fixed

We tested whether known command frequencies can improve stimulation-site prediction sets. We used 171 sites and 342 test pulses from five public SEEG sessions. Each simulated command mix assigned 80 percent of commands to one shaft. The response distribution at each site stayed fixed. Across all 38 shaft policies, the patient mean of minimum correct-site coverage rose from 63.5 to 89.8 percent with frequency weights. Four weighted policies still had coverage below 90 percent. Mean candidate count across policies was 10.31, but it grew at the policies with lowest original coverage. In 16 policies, the small calibration sample forced all favored-shaft sites into the set regardless of voltage. The package supplies a cutoff tool, compact data, source hashes, figures, and reproduction code. A fresh raw rebuild and a separate repeat matched exactly. This exploratory within-session result does not give a coverage guarantee for future pulses or live BCI use.

stereoelectroencephalography · prediction sets · label shift · coverage · electrical stimulation · brain-computer interfaces · calibration