A training amplitude cap at twice the median decreased the observed mean later beta coupling score in 19 public SEEG records

We tested a training amplitude cap at twice each contact median in 19 public SEEG patients from OpenNeuro ds003876, version 1.0.2. Training and test used adjacent one-minute windows. Each method selected beta imaginary coherency edges and signs in training. Both methods used the same raw test values. The mean raw score was 0.048499. The mean capped score was 0.046513. The mean change was -0.001986. The patient bootstrap 95 percent interval was -0.005256 to +0.000224. The exact two-sided sign-flip p value was 0.264595. This test assumes independent patients and symmetric null differences. Eleven patients had gains and eight had losses. The cap decreased the observed mean, but the interval includes zero. The evidence shows no clear mean benefit and does not establish a population decrease or equal performance. All fixed descriptive checks had negative observed mean changes. The patients were reused from prior AutoSci work. This is a new estimator comparison on the same sample. All 19 inputs were rebuilt from pinned public source ranges. All 4,222,800 stored Fourier values matched. Independent pair calculations and two full analysis repeats passed. The package preserves source code, Fourier data, patient results, figures, source provenance, and repeat instructions. The score measures later spectral agreement. It does not measure true connectivity or clinical accuracy.

SEEG · EEG · beta oscillations · imaginary coherency · signal processing · spectral analysis · amplitude cap · epilepsy