Bin shuffles rejected 24.3 percent of uniform circular audio controls at a 5 percent threshold in five ds004703 SEEG patients

We tested two reference rules for speech and SEEG correlations. We used 23 later speech passages from five patients in OpenNeuro ds004703 version 1.1.0. An earlier study chose each contact pair and lag from early passages. We kept those choices fixed. Each control independently rotated the cropped audio vector within each later passage. Rotations were uniform over the complete finite cyclic set. Each patient supplied 20,000 independent controls. Each rule used 9,999 independent reference records and a two-sided 5 percent threshold. Bin shuffles rejected 24.316 percent of controls on average across patients. Circular references rejected 5.005 percent. The difference was 19.311 percentage points. Exact circular Pearson variance was 2.945 times the bin-permutation variance on average across patients. The difference persisted at zero lag, after linear trend removal, and with a second seed. This is a conditional control benchmark. It does not prove that circular rotations are valid for every natural speech process. It reuses patient data and makes no new population or causal claim. Fresh public data reproduced all 23 selected later feature pairs exactly. Independent calculations and a clean offline rerun passed. The package preserves code, features, controls, proof, figures, source versions, hashes, and reproduction steps.

SEEG · speech envelope · circular rotation · bin permutation · calibration · correlation variance