At a zero threshold, eight-pair SEEG decoders gave a mean of 36.7 outside-guard triggers per eligible minute in eight later task scans

We tested whether selected rest windows describe the output of a continuously applied SEEG decoder. We used OpenNeuro ds004473 version 1.0.2 and eight patients. The first 40% of each retained trial list fit an eight-pair ridge model. The next 20% set two rest thresholds. Full intervals around the last 40% gave 112.3 minutes and 67,309 decisions. At the fixed zero threshold, mean selected-window balanced accuracy was 72.2%. Positive rates were 27.6% in selected rest and 28.3% in scan windows fully outside one-second response guards. The mean paired change was 0.7 percentage points (95% person bootstrap interval -3.0 to 4.3; exact sign-flip p=0.7109). A one-second command spacing rule gave 36.7 outside-guard triggers per eligible minute. The 99th early-rest percentile rule gave 7.8 triggers per eligible minute, with 23.8% pre-response trigger detection versus 47.6% at zero. These triggers are a proxy for unwanted commands. Other movements were not measured. All tests were offline and within one session. The package preserves all scores, models, features, source versions, range hashes, figures, and reproduction code.

stereo-EEG · brain-computer interface · continuous decoding · false command rate · ridge regression · neural signal processing