Fitted residuals had positive mean correlations in six iEEG patients and two story panels from 24 fMRI adults

We tested whether sound and recent-signal models remove response timing shared across listeners. Separate public studies supplied human iEEG amplitude and BOLD fMRI. Fixed references used two patients and six MRI adults. Development used one other patient and four other adults. We added the residual-correlation claim after development. A local code lock preceded tests in six other patients and 24 other MRI adults. The same MRI adults heard two stories. Mean fitted residual correlations were 0.07307, 0.11324 and 0.10715 for iEEG, Merlin and Sherlock. Their pointwise 95% person Student limits were [0.03024,0.11589], [0.07754,0.14894] and [0.07015,0.14416]. The two-claim shared Holm p was 0.03125, under independent targets and null sign symmetry. The original full-reference prediction-gain test failed. Mean gains were +0.6909, -2.6316 and -2.7004 R2 points. All negative people and supporting checks remain. An exact conditional selection example gave residual correlation 0.258199 from independent zero-response test signals. Thus the human result concerns fitted errors, not an identified missing neural component. Fixed references and stories, reused selection, clinical coverage, full-record cleaning, cohort differences and vascular filters limit scope. The package gives all numerical inputs, code, source pins and terms, predictions, residuals, uncertainty, figures and repeat instructions. It establishes no common neural cause or causality.

human neuroscience · iEEG · fMRI · intersubject correlation · residuals · prediction · selection bias