With 32 calibration sounds, acoustic prior transfer failed a shared test in 20 iEEG patients and 16 fMRI adults

We tested whether acoustic coefficient priors transfer between separate human iEEG and fMRI studies. A local lock fixed a 32-sound test in 20 patients and 16 MRI adults. The shared gain-and-skill rule failed, p=0.088217. Mean gains were 0.467% in iEEG and 2.433% in MRI. A planned 64-sound check later passed a Holm correction across 15 diagnostic propositions, p=0.010176. Mean gains were 0.986% [0.436,1.536]% and 0.334% [0.170,0.499]%. Brackets give pointwise 95% person Student limits. This inference is exploratory and reuses the primary people. The 64-sound MRI gain did not clearly exceed shuffled source references; its within-category reference fraction was 0.700. Its person/common-sound gain range also included zero. Thus no shared neural tuning benefit was confirmed. A separate eight-adult MRI study gave a positive 32-sound gain. The method uses 86 documented acoustic features, ten common components, source coefficient covariances, and separate target penalty selection. It needs no match of people, sites, or sounds between electrical and MRI studies. An exact noise limit shows that source acoustic design alone can give a non-isotropic prior. All negative people, controls, and a one-factor limit model remain. This is a tool and benchmark result for fixed libraries. It does not establish a common neural cause, equal signals, broad new-sound transfer, or causality. The package has code, permitted inputs, source pins and terms, all outcomes, figures, and repeat instructions.

human neuroscience · intracranial EEG · fMRI · auditory cortex · encoding models · covariance prior · transfer learning · negative result