Area scores from four axes were positive for 165 sounds after class mean removal in six SEEG patients and 24 fMRI adults

We tested if natural sound responses contain a repeatable second linear direction after three voice class means are removed. A weighted four-axis score uses one repeat to choose axes and two other repeats to estimate response area. A common rank-one mean has zero expected score when repeat errors are independent and have zero mean. Noise within a repeat can have any covariance. Separate public studies supplied SEEG high-gamma activity and BOLD fMRI. Development used two patients and four adults. A local code lock preceded new endpoint tests in six other patients, 16 other adults and eight adults from a separate fMRI study. These are reused cohorts and local endpoint holdouts. All 30 reserved people had positive scores. Mean scores were 0.006636, 0.010757 and 0.007368. Their 95% person t intervals were [0.003584,0.009687], [0.008154,0.013360] and [0.003592,0.011144]. Both fixed shared rules passed a two-claim Holm correction, p=0.002529. These are measured score results for fixed source panels. Source selection used repeat data, so the conditional rank theorem does not certify latent neural rank here. An exact example gives score +0.09 after repeat-based selection despite a rank-one unconditional mean. A later electrical baseline subtraction kept a positive mean but changed the score scale. It used the same six patients. A separate solver and original source extraction checked the human results. A final search found a concurrent three-repeat minor assay, which we cite. This work gives a weighted extension and a sharp repeat-count rule for conditional multilinear tests. Four total repeats are needed for its adaptive rank-two test; no fourth human repeat is available. The package preserves code, all scores, model failures, source versions and terms, local plans, figures and repeat steps. Clinical sampling, separate cohorts, unequal coverage, source processing and Student assumptions limit scope. No common neural cause is established.

SEEG · fMRI · auditory cortex · repeat reliability · rank-one mean · response geometry · source selection · cross-validation