A fixed sound panel test did not confirm shared gains in 20 iEEG patients and 16 fMRI adults

We tested whether acoustic features can select a small sound panel that lowers neural prediction error. Separate public human studies supplied iEEG voltage and fMRI BOLD. Development used four people per study and separate sound identities. A local lock fixed a 16-sound panel, kernel settings and two comparisons before reserved tests. In 20 other patients and 16 other adults, mean gains over random panels were -0.093% and +4.155%. Pointwise 95% person Student intervals were -2.635% to +2.449% and +3.388% to +4.922%. Gains were positive in 9/20 and 16/16. Both shared tests failed; Holm p was 1. A person and common-sound resample crossed zero in both measurements. The fMRI comparison with diversity selection stayed positive under that resample, but its electrical comparison failed. An exact later decomposition gave constant-mean parts of +0.990% and +5.415%, larger than the total means. Thus a panel gain alone does not show a better acoustic response function. Negative values and all fixed checks remain. An independent solver and direct errors checked all 36 primary person records and 260 panel prediction maps. The contribution is a new retrospective benchmark, not a new selection algorithm or new-study replication. Source selection, distinct cohorts and sounds, weak voltage prediction, trial averaging and unequal measurement scales limit scope. No positive cross-modal neural or causal claim is made. The package gives code, permitted electrical means, all errors and panels, plans, source hashes and terms, figures and repeat steps.

auditory cortex · experimental design · intracranial EEG · fMRI · natural sounds · Gaussian process · stimulus selection · negative result · reproducibility