A one-second spatial fit gave a lower mean 70–110 Hz test score than channel selection in 21 SEEG participants

We tested whether a one-second spatial fit gave a larger repeated power change than selected-channel power. We used all 35 released SEEG records from 21 people in OpenNeuro ds006519, version 1.0.0. A fixed panel kept at most eight bipolar channels outside the stimulation shaft. The primary spatial fit used a 50% diagonal blend and 32 retained bins within 70–110 Hz. Both methods used baseline and marked post seconds to fit, then separate seconds to test. Records were averaged within each person. People had equal group weight. The spatial method gave a larger mean absolute fit contrast, 8.619 versus 6.513 dB. Its mean signed test score was lower, 2.025 versus 2.509 dB. The mean difference was -0.484 dB. The 95% person bootstrap interval was -1.069 to +0.048 dB. The exact person sign-flip reference gave p = 0.103 under its stated symmetry assumption. The interval includes zero and does not establish a population difference or equal performance. All fixed non-diagonal post variants had negative observed mean differences. The full diagonal control exactly matched channel selection. The package contains exact native samples, matrices, weights, figures, code, provenance, and offline replay instructions. Independent reader checks, raw source hashes, a second extraction, and an exact numerical replay passed. These scores measure recorded power change. They do not measure command decoding or a causal effect of stimulation.

SEEG · intracranial EEG · spatial filtering · common spatial patterns · power repeatability · brain–computer interfaces