A fixed visual stage contrast did not pass a conservative transfer test in human MEG and fMRI studies of 92 and 118 images

We tested whether a change in visual pair relations agrees between human MEG and fMRI after each image's overall distance effect is removed. Development used 92 images in 16 people with MEG and 15 with fMRI. A fixed test used 118 images in 15 people with both modalities in a separate study. Both modalities shared people within each study. Cross-modal person links were absent from the released targets. We used conservative bounds over all possible ten-person test sets outside five reference people. Full MEG matrices supplied the same 70-90 and 140-160 ms windows in both studies. The second study's fusion peaks did not set its windows. The stage interaction passed in development, shared p=0.001953, but failed in confirmation, shared p=0.989258. Confirmation mean interactions across all rows were 0.00453 in MEG and 0.01046 in fMRI. Outer conditional person limits were -0.02345 to 0.03654 and -0.00155 to 0.02081. A stricter crossover check failed in both studies. Rank and finite visual controls retained the shared failure. A model with unchanged latent geometry and altered noise produced positive interactions in both directions. Thus a positive interaction alone need not show a new representation. The package gives the failed fixed test, the model guard, all signed results, a labeled exploratory time course, source hashes, code, figures and repeat steps. Original arrays remain external. Different cohorts, stimuli, protocols, low repeatability and fixed reference people limit scope. No causal claim or positive cross-modal transfer claim is made.

human vision · MEG · fMRI · representational similarity · independent confirmation · negative result · measurement reliability · streaming analysis