Button side predicted face class within ds000117 v2.0.0 sessions in 16 people. Across separate identities, fMRI to MEG transfer gave 42.4% balanced accuracy.

This study tested famous versus unfamiliar face labels with public button events from 16 people in ds000117 version 2.0.0. Five fixed identity folds kept training and test identities separate. Each person had a rule that used only button side. Mean balanced accuracy was 58.3% within fMRI and 61.6% within MEG. Their four-condition adjusted upper-tail permutation values were 0.0002 and 0.0001. The primary fMRI to MEG test gave 42.4% accuracy, with a 37.96–47.19% person bootstrap interval. Its planned upper-tail permutation value was 1.0. It did not show transfer above 50%. The full-session response association changed sign in 12 of 16 people. The cause is unknown. With the fitted rules fixed, all completions of excluded MEG responses gave mean bounds of 40.76–44.34% in this cohort. The package preserves 240 event tables, source hashes, code, response links, folds, predictions, random draws, results, figures, and reproduction instructions. Independent calculations and complete reruns matched the saved results. No brain signal was tested.

button side · face class · fMRI · MEG · response bias · identity folds · balanced accuracy · OpenNeuro · ds000117