A donor sound direction did not pass a shared prediction gain test in six iEEG patients and 20 fMRI listeners after 180 s fits

We tested whether a sound response direction from other people improves short individual fits in two human neural measures. Separate public studies supplied nine iEEG patients and 25 fMRI listeners. Fixed donor groups had three patients and five listeners. The test used six other patients and 20 other listeners. Each target model used 180 seconds for coefficient fitting and predicted a later speech segment. A donor direction was left free of the ridge penalty. Controls used ordinary ridge and reversed donor delay weights. Mean test R2 gain over ordinary ridge was 0.001464 in iEEG and 0.133153 in fMRI. Their 95% person bootstrap intervals were 0.000102 to 0.002935 and 0.099434 to 0.168570. The iEEG Student t interval included zero. Exact mean sign-flip p values were 0.09375 and 0.000000954. Gains over the reversed direction were positive in all six and all 20 test people. The shared primary rule required both gains in both measures. It failed, with conjunction p 0.09375. The mean iEEG gain over ordinary ridge became negative in the second test half. A planned 60-second fMRI check was infeasible and is retained as a failure record. These reused cohorts provide a fixed new-endpoint test, not independent new-cohort confirmation. Offline cleaning, inherited contact selection, shared stories, clinical sampling and unequal spatial and temporal scales limit scope. The package preserves compact inputs, source versions and terms, code, plans, coefficients, all person and output results, figures, audits and repeat instructions. The contribution is a donor-transfer benchmark and a measured limit on a shared gain claim. It does not establish a common neural cause.

human neuroscience · intracranial EEG · fMRI · speech · transfer learning · ridge regression · negative result · cross-modal test