Rolling error bounds did not establish a lower mean interval score for response time in eight final SEEG task blocks

This package gives a response time interval benchmark for brain interface research. It uses public OpenNeuro ds004473 version 1.0.2. The test had eight patients and 548 trials in final task blocks. Earlier trials fit each model. The next 32 trials per patient set 90% bounds. Static bounds kept the calibration errors. Rolling bounds used the last 32 completed errors. Coefficients stayed fixed. The primary equal-patient static-minus-rolling log interval score was 0.1005, with a descriptive 95% patient bootstrap interval of -0.0067 to 0.2693. The exact sign-flip p value was 0.2656. This test did not establish a lower mean score. The sign-flip test assumes independent patient effects symmetric about zero under the null. Rolling history bounds reduced mean absolute coverage error from 7.38 to 3.27 percentage points, a secondary result. The specified pre-cue SEEG power correction increased the rolling score in all six patients with a nonzero correction. This reused clinical cohort does not test live control or brain stimulation. The package contains code, input features, source pins, all test decisions, three figures, an independent raw audit, and reproduction commands.

response time · prediction intervals · conformal prediction · stereoelectroencephalography · brain-computer interfaces · calibration · interval score