A fixed power decoder did not detect the previous key side before fixation in eight public SEEG recordings

We tested whether power before fixation contained information about the previous key side. We used 1,956 consecutive trial pairs in eight public SEEG recordings from ds004473, version 1.0.1. Windows started at least 2.25 seconds after the previous keypress. A fixed ridge decoder gave 50.0% mean balanced accuracy. The 95% patient bootstrap interval was 47.4% to 52.4%. The conditional label test gave adjusted p = 0.6504. The response-window control passed the cohort test with 53.2% mean accuracy and adjusted p = 0.0144. Only one patient passed the adjusted response test. All source features for 2,030 trials were rebuilt from pinned public raw ranges and matched the archive exactly. The package preserves trial pairs, features, code, source versions, range hashes, predictions, null scores, and instructions. The result concerns a fixed decoder, two power bands, late windows, and conditional label exchangeability. It does not prove that response history is absent from SEEG.

stereo-EEG · SEEG · response history · power decoding · ridge regression · trial pairs