Trial self-product correction reduced mean calibrated beta interaction scores at keypress by 95 percent in eight SEEG patients

We tested beta interaction scores near a keypress after removal of products from the same trial. The source was OpenNeuro ds004473 v1.0.2, with eight patients and 64 selected trials per patient. The main analysis used 32 calibration trials and 32 separate evaluation trials. It tested 1,281 path pairs across nine beta frequency bins. Correction reduced the mean keypress score from 0.0140912 to 0.000663175, or 95.3 percent. It changed 67 of 130 places in the patient top response edge lists. The corrected response minus baseline mean was 0.000441879. The patient bootstrap 95 percent interval was -0.000138088 to 0.00119779. The exact patient sign-flip p-value was 0.296875 under joint sign symmetry. All planned sensitivity intervals also included zero. The estimator needs independent common-mean trials and independent calibration for its unbiased interpretation. The data selection does not establish these assumptions. The score correction is an exact data identity, not a measured fraction of artifact. Simulations test known targets and show failure with serial trial dependence. Scores agreed across two real bases of the fixed contact path spaces. The package includes code, Fourier inputs, full scores, a proof, figures, source locks, and rebuild instructions. A fresh range rebuild exactly matched all 1,134,720 complex Fourier coefficients. An offline repeat and file hash checks passed.

SEEG · intracranial EEG · beta interaction · trial self-products · cross-moment · variance correction · U-statistic · signal processing