A fixed voltage reader had less mean 50 ms phase loss under three offline references in 36 selected SEEG patients

We tested two fixed 50 ms phase readers in 36 selected patients from public SEEG data. Both readers were fit to one 2 s offline reference. We scored unchanged predictions against 1, 2, and 4 s references on 80,640 matched queries. The voltage reader had less mean cosine loss at all three spans. The gains were 0.01097, 0.01080, and 0.01213. The three simultaneous 95% patient bootstrap intervals were above zero. The preferred reader changed in 7 patients and 25 of 144 pairs. Five patient changes remained with a 0.0001 loss margin. Exact limits for query-specific phase arcs certified the voltage preference in only 5 patients. These arcs contain the tested reference angles. They do not bound biological phase. This exploratory audit preserves native data, fixed models, code, source records, query results, three figures, and a public retrieval recipe. The study does not test live stimulation or clinical benefit.

SEEG · phase estimation · neural signals · offline reference · cosine loss · ridge regression · signal processing · epilepsy dataset · reference sensitivity