A test that removed 69 acoustic features did not confirm sound agreement in one iEEG panel and eight fMRI adults

We tested whether sound response patterns agree across separate human iEEG and fMRI studies after full linear feature control. Public derivatives supplied 32 distinct selected electrode profiles and scan 3 from eight fMRI adults. All studies used the same 165 sounds. The electrical patient map was not released, so the panel was one fixed reference. We removed the response span of 69 supplied acoustic features and 12 categories. We then removed every kernel direction that independent, unequal sound error variances can produce after that projection. Mean corrected cosine was 0.015586, with 95% person t interval -0.000179 to 0.031352. Six of eight adults had positive scores; exact sign p was 0.144531. The fixed rule failed. A person bootstrap range excluded zero, and earlier BOLD scans gave larger scores; all discordant results remain. In 500 model draws, independent unequal-variance errors had mean projected cosine 0.5055 before the error correction and -0.0025 after it. Correlated sound errors survived the correction. Full feature removal can also discard true signal. The tool implements an established projection principle and recovers U-centering for an intercept. This is a fixed new endpoint in reused data, not new-cohort confirmation. It does not establish shared non-acoustic coding, its absence, or causality. Missing electrical person labels, selected tissue, fixed sounds, source processing and incomplete acoustics limit scope. The package preserves code, source hashes and terms, all scores, model checks, figures, plans and repeat steps. Original source matrices remain external.

auditory cortex · intracranial EEG · fMRI · representational similarity · acoustic features · measurement noise · negative result