Independent contact models of SEEG spectra from 12 patients gave unequal ten-sample prediction gains in shared-contact bipolar pairs

We tested shared bipolar subtraction with public SEEG spectra from 12 patients in OpenNeuro ds004100 version 1.1.3. Each patient supplied two one-minute records. We kept the processed contact spectra and imposed contact independence at every lag. The primary predictor used ten past samples per channel at 100 Hz. The mean patient median absolute difference between directions was 0.001438. The descriptive 95 percent patient bootstrap interval was 0.001049 to 0.001878. The mean patient fraction above a 0.01 net-gain threshold was 3.80 percent. Disjoint pairs and equal normalized spectral shapes gave numerical zero. Equal contact variances did not remove the effect. An exact example, sample regressions, time reversal, and a separate Cholesky calculation checked the method. A fresh raw rebuild matched both HUP060 data archives. A repeated analysis matched all six numeric files. The added scope is a 12-patient finite-history spectrum benchmark for an established mixing effect. Independence was imposed; the results do not measure neural connections or a patient artifact rate. The package preserves code, periodograms, source hashes, proofs, results, figures, and reproduction instructions.

stereoelectroencephalography · SEEG · bipolar subtraction · contact overlap · prediction gain · spectral model · signal mixing · time series