With one fixed penalty, acoustic sound selection failed a shared test in 20 iEEG patients and 16 fMRI adults

We tested whether acoustic features can choose 32 sounds that improve prediction of neural response differences. Separate public studies supplied iEEG LFP and fMRI BOLD. Five patients and four adults supplied development. A local code lock fixed one penalty before tests in twenty other patients and sixteen other adults. A greedy linear design optimized mean-free acoustic contrasts. Random reference sets had the same sound category counts. Test means were removed before error scoring, so constant forecast offsets could not give contrast gains. Mean gains over random sets were +0.114% and +2.911%. Their 95% person Student intervals were [-0.615%, +0.843%] and [+2.615%, +3.207%]. Positive counts were 10/20 and 16/16. Selected contrast skills over a constant forecast were +3.396% and +15.449%. The fixed shared gain-and-skill rule failed, p=0.58809853. Person and common-sound checks retain fixed fits. A fixed uniform-random check gave MRI mean gain -0.937%. All negative outcomes remain. Nominal model-variance reductions did not ensure a shared measured gain. The package gives a new benchmark, not a new selection algorithm. It preserves code, licensed electrical inputs, source pins and terms, primary scalar errors, moments, all checks, figures and repeat steps. Separate cohorts and sounds, sparse clinical sampling, source screens, trial averaging, unequal features and sensing, and one common penalty limit scope. No positive shared selection benefit or common neural cause is established. Original MRI arrays and audio remain external.

human neuroscience · intracranial EEG · fMRI · auditory cortex · stimulus selection · optimal design · replication · neural prediction