Separate sound band histories failed a shared prediction test in seven iEEG patients and 28 fMRI adults

We tested whether a separate one-second history for each sound-frequency band improves later brain-response prediction. Public Podcast iEEG and Narratives fMRI supplied separate studies and cohorts. Development used two patients and six MRI adults. Fixed tests used seven other patients and 28 other adults, with two stories per MRI adult. The baseline had current spectra, past spectra, broadband history, gain terms and word timing. Training-only feature projection and a fixed baseline prediction controlled changes in baseline shrinkage. Mean added test R-squared was -0.00000798 in iEEG, -0.00034368 in Merlin and -0.00662052 in Sherlock. All pointwise 95% person t intervals crossed zero. Positive gains occurred in 2/7, 3/28 and 3/28 people. The shared positive rule failed, sign p=0.999998484. Validation chose no added prediction in 3/7, 19/28 and 17/28 records. We retain all negative results and fixed controls. Fresh source checks matched 204 MRI objects and one complete electrical input reconstruction. An independent loss audit checked all 77 main records. The contribution is a controlled negative result and a reusable assay. It does not establish absent adaptation or positive cross-modal neural evidence. Clinical sampling, separate stories, regional MRI averaging, measurement filters and finite-sample projection limit scope. The package gives code, permitted inputs, source versions and terms, results, figures, local plans and repeat steps.

human neuroscience · intracranial EEG · fMRI · auditory context · spectral history · held-out prediction · negative result · reproducible assay