Eight learned sound response directions failed a shared gain test in six SEEG patients and 24 fMRI adults

We tested whether a few response directions improve prediction of repeated human sound responses. Separate public studies supplied SEEG high-gamma amplitude and two BOLD fMRI groups. Two patients and four MRI adults supplied development. This selected eight directions and a prediction factor of 0.25. A local code and input lock preceded tests in six other patients, 16 B2021 adults and eight NH2015 adults. Other sounds and one repeat fitted the directions. Two other repeats supplied the input and target for test sounds. Class means and spatial means were removed. The comparator copied the input with a factor set in development. Mean PCA gains over zero were +4.596, +6.127, +5.040 points of observed residual variation. Mean added gains over uniform shrinkage were +1.042, -0.099, -0.004 points. The shared p was 1. The fixed shared test failed. Added value over simple shrinkage was not confirmed across the three studies. All negative scores and fixed checks remain. Short-baseline subtraction changed SEEG gain over zero to -0.933 points. This check changes the target and noise; it does not prove an artifact. Source selection used the full sound panel. Separate cohorts, sparse clinical coverage, common schedules, three repeats and unequal measurement filters limit scope. The score does not identify intrinsic dimension, a common neural cause or causality. The package supplies code, source pins and terms, all scalar results, uncertainty, figures, audits and repeat steps. Original neural arrays remain external.

human neuroscience · SEEG · fMRI · auditory responses · PCA · shrinkage · prediction · cross-modal test