Four pre-pulse coefficients increased mean relative SEEG prediction error in five patients at 50–250 ms with ridge penalty ratio one

We tested signed voltage detail before stimulation. The task was centered SEEG shape prediction beyond local amplitude and an early site mean. OpenNeuro ds004457 version 1.0.2 supplied five epilepsy patients. The ridge model used four signed cosine coefficients. Its penalty ratio was one. Slopes came from other sites in each patient. Five early pulses calibrated each site. Five later pulses supplied 50 to 250 ms test windows. There were 1080 scored links. Mean relative after-pulse error increased 0.797 percent. Mean relative before-pulse error increased 0.362 percent. The paired gain difference was -0.436 percentage points. Descriptive patient resample limits were -1.457 to 0.189 points. The conditional patient sign-flip probability was 0.6875. All seven planned checks had negative paired differences and ranges that included zero. Mean physical after-pulse error fell slightly. This change has different weighting from mean relative gain. The test did not establish a stimulation-specific prediction benefit. It did not establish absence of useful state information. All 120 raw files passed pinned source checks. A raw rebuild and analysis repeat reproduced all saved arrays and tables. Independent least squares checked 4320 primary fits. A second NumPy build reproduced all numeric results within a fixed tolerance. The package preserves code, matched input windows, all scores, primary fits, predictions, source hashes, and repeat instructions. This is an offline within-patient test with local site calibration. It does not test new patients or causal connectivity.

stereoelectroencephalography · electrical stimulation · voltage prediction · ridge regression · matched control · prediction error