Separate state sums preserved SEEG site decisions during known invertible reference switches in five offline sessions

A known software reference change can occur inside a SEEG voltage-averaging bin. We tested separate sums for each reference state on OpenNeuro ds004457 version 1.0.2. The test used five offline sessions, 183 stimulation sites, and 732 test pulses. At the planned 30 ms switch, separate sums preserved all fixed-reader decisions and scores exactly. Correction of one pooled mean changed 34 of 732 decisions and caused bin errors up to 43.84 microvolts. Four sessions had a nonidentity change. The fifth was a one-pair control. Separate sums preserved all 24,888 condition-pulse decisions across 34 planned schedules. The schedules reused the same pulses. A proof and a two-cell counterexample explain the result. Fresh raw extraction matched every archived input array and event table. The package preserves code, source hashes, scores, figures, and rebuild instructions. The result requires known invertible software transforms at source-cell boundaries. It does not test physical switching artifacts or clinical benefit.

SEEG · intracranial EEG · reference switching · voltage averaging · state-sum recovery · signal processing · brain-computer interface