A fixed early power rule gave no clear late stability gain with separate references at about 80% retention in 99 SEEG records from 19 people

We tested whether early post-stimulation power can improve later channel selection beyond a baseline rule. We used public SEEG data from OpenNeuro ds005169, version 1.0.0. The primary test used 99 records from 19 people and 9860 channel-record pairs. Both rules kept about 80% of channels. The early score and late outcome used separate reference windows. The equal-person late large-change rates were 10.24% after baseline selection and 9.81% after early selection. The gain was 0.43 percentage points. Its 95% person bootstrap interval was -0.65 to 1.39 points. The exact sign-pattern test gave p = 0.439. A planned shared-reference check gave a larger gain of 2.23 points. A post hoc reference swap also changed the estimate. These results show reference dependence. They do not identify its cause or prove equal performance. Power stability does not measure decoder accuracy or useful neural responses. The package preserves 112 compact spectral records, source hashes, code, plans, complete channel decisions, tables, figures, and reproduction steps. Two fresh raw checks matched every saved array exactly. An independent calculation and a byte-exact replay reproduced the main results.

SEEG · channel selection · power stability · electrical stimulation · brain-computer interfaces