Two fixed sound volume tests failed a shared SEEG and fMRI rule; repeat screening biased a rank-one control

We tested third-order geometry among 128 sounds outside source human-voice groups. Separate public studies supplied human SEEG high-gamma responses and BOLD fMRI. Each person had 24 sites. We averaged all four-sound subsets and used a third repeat for site scales. A prior Gaussian correction addressed correlated errors under equal independent Gaussian covariance. Development used two patients and four MRI adults. A local lock preceded six other patients and 16 other adults. Raw means were +0.01896 [-0.00012,+0.03804] and +0.04890 [+0.03189,+0.06592]. Corrected means were -0.27070 [-1.00165,+0.46024] and +0.01183 [-0.03658,+0.06024]. Brackets give 95% person Student limits. Neither shared claim passed the mean, majority and interval requirements. Holm p values were 0.05100 and 0.80759. Eight adults from another MRI acquisition supplied a check. It did not rescue the shared claims. A known rank-one simulation tested selection by repeat reliability. Its corrected mean was +0.007149 with test-repeat selection and -0.000021 with independent selection. In an added exact rank-one model, a repeat-agreement screen gave positive corrected expectation over a stated threshold range. These results extend prior volume methods to a larger panel and define a selection failure for the correction. They do not establish absent sound information or a shared neural cause. Source screening, unverified noise assumptions, sparse clinical sampling and unequal measurement filters limit interpretation. We preserve code, all outcomes, source pins, figures, audits and repeat instructions. Original neural arrays remain external.

human auditory cortex · SEEG · fMRI · response geometry · noise correction · selection bias