Fixed sound dictionaries gave mean expected random label accuracy above chance in six SEEG patients and two fMRI studies

We tested whether fixed sound dictionaries can carry random labels across people in human SEEG and fMRI. Separate studies supplied high-gamma electrical responses and BOLD response estimates. A local code lock followed development in two patients and four MRI adults. Each study used 32 calibration sounds and 96 local queries, with 48 candidate templates per query. We computed exact expected accuracy over all balanced random binary labels. Mean accuracy was 53.01% in six reserved patients, 58.84% in sixteen reserved B2021 adults, and 58.71% in eight NH2015 records tested with the same MRI dictionary. The 95% person Student limits were 51.33-54.70%, 58.03-59.66%, and 57.17-60.25%. Chance was 50%. One shared claim required person mean, exact sign-flip and common-identity tests in all panels. It passed, p=0.015625. A disjoint-identity gallery of the same size has exact expected accuracy of 50% under independent half codebooks; this is a mathematical reference. A short trial-baseline subtraction left the SEEG interval across chance. All controls and unfavorable codebooks remain. A two-pattern model also gives positive expected transfer, so the result does not identify unique item or semantic codes. Cohorts, sampling, source screens, calibration and vascular filters limit scope. MRI cross-study person overlap could not be audited. No direct SEEG-to-MRI clip transfer or cause was established. The package gives source pins and terms, code, statistical summaries, independent checks, figures and replay steps.

SEEG · fMRI · sound identity · random labels · subject generalization · stimulus generalization · neural decoding · reproducibility