Correct trial matching did not show lower later probability error for conditional logistic readers in eight SEEG recordings

We tested whether correct trial matching reduced later probability error in public SEEG data. The study used eight people, eight selected bipolar pairs per person, and 674 later test trials. Weights and probability calibration used separate early trials. The primary Brier error was 0.1911 for correct matching and 0.1899 for the mean of 100 random-match readers. The mean difference was +0.00125, with a 95 percent person bootstrap interval [-0.00179, +0.00487]. The two-sided sign-flip value was 0.570. This test did not show a benefit. It did not establish equal performance. A paired fit cancels shared additive feature offsets when preprocessing stays fixed. This proof does not give a guarantee for separate window predictions. All features were rebuilt from pinned public source ranges. The package includes data, code, models, figures, provenance, and repeat instructions. This is an exploratory offline study of one recording per person. It does not test brain stimulation. This package revision excludes sci download metadata during file verification.

stereoelectroencephalography · conditional logistic regression · trial matching · probability error · brain-computer interfaces