A four-class area test failed its shared rule in 20 iEEG patients and 16 fMRI adults

We tested whether human-voice and music contrasts need more than one linear response direction in two separate human studies. Four development people per modality preceded a fixed test in 20 other iEEG patients and 16 other fMRI adults. Four disjoint sound groups defined two contrasts. A correction removed the expected squared-area term from finite clip sampling under independent class samples and fixed calibration. Mean corrected area was 0.030763 in iEEG and 0.209651 in BOLD. Person 95% Student intervals were [0.014559, 0.046967] and [0.142542, 0.276760]. Only 13 of 20 patients had positive scores, compared with 16 of 16 adults. The fixed shared mean, sign and fitted-reference rule failed, p=0.131588. A centered person-and-clip range included zero in iEEG. All negative scores remain. A later exact derivation gives a site-count rule after spatial mean removal: F(p)=p squared divided by (p-1)(p-2), and the mean of F(k) times the k-site score over uniform subsets equals F(p) times the full score. Every electrical three-site subset verified the identity to 3.41e-16. This does not model clinical electrode placement. Acoustic residual scores were positive on average, but calibration noise alone could create residual area in a synthetic example. Neither residual area nor raw area counts neural causes. The package gives a small-memory tool, proof, permitted inputs, source pins and terms, all outcomes, figures and repeat instructions. Unequal cohorts, clips, spatial coverage, preprocessing and source selection limit scope. No positive shared biological or causal claim is made.

human neuroscience · intracranial EEG · fMRI · auditory cortex · representational geometry · unbiased estimation · site sampling · negative result