A quadratic model failed a shared sound prediction test in 20 iEEG patients and 16 fMRI adults with whole groups excluded

We tested whether a fixed quadratic acoustic model predicts sound differences inside a broad source group excluded from fitting and tuning. Separate public human studies supplied signed intracranial LFP means and BOLD fMRI. Five patients and four MRI adults supplied development. A local code and input lock preceded tests in 20 other patients and 16 other adults. Both models used 88 bounded acoustic summaries. The augmentation added an eight-coordinate quadratic kernel. Mean gains were -0.240 and -1.395 points of class-centered null error. Their 95% person Student intervals were [-0.997,0.517] and [-2.338,-0.452]. Positive counts were 5/20 and 3/16. The fixed shared gain-and-skill rule failed, p=0.99790955. The MRI gain range crossed zero when fixed-fit resampling also varied common sounds. A descriptive mixed-fold MRI check gave +3.806 points in all 16 adults. An added eight-person MRI study gave an uncertain +0.127-point mean on the same sounds. Negative and discordant results remain. Specified quadratic controls gave gains in all 24 draws per acoustic panel. Independent direct calculations checked 360 selected outer fits and 640 inner candidates; a compact repeat recovered every primary score and choice exactly. The contribution is a whole-group transfer benchmark, compact sound statistics and checked code. Clinical coverage, source screens, separate libraries and cohorts, unequal resolution, finite model choice and fixed-fit uncertainty limit scope. No positive neural generalization across modalities, universal curvature cost or common cause was established. Source versions, terms, provenance, all results, figures and repeat steps are supplied.

human auditory cortex · intracranial EEG · fMRI · sound encoding · quadratic kernel · group generalization · negative result · reproducible benchmark