Fixed tests did not confirm a shared tree compression loss in late MEG and IT fMRI across two human image studies

We tested whether average-linkage tree compression loses image-distance relations that predict other people. Public human MEG and IT BOLD fMRI supplied a 92-image development panel and separate 118-image and 78-image confirmation experiments. Each modality used five reference rows and ten other target rows. No cross-modal person join entered the test. Two fixed shared claims tested raw rank maps and maps after image and broad-class control. Both failed, Holm p = 0.906238. Adjusted mean full-minus-tree correlations were 0.0885 and 0.0165 in the 118 MEG and fMRI panels, and 0.0483 and 0.0081 in the 78 panels. The 78 IT difference interval was [-0.0160, 0.0321]. All negative people and controls remain. A direct pattern model with one identical true image pattern gave adjusted loss 0.4494 from image-dependent error size alone. This is an interpretation limit, not a fitted human cause. The contribution is a checked compression benchmark, a complete negative cross-modal test, an exact measurement example, code, source pins and terms, figures and repeat steps. Source arrays stay external. Fixed images, small references, source filters and unequal signal precision limit scope. No positive shared neural geometry or causal result was confirmed.

human vision · MEG · fMRI · hierarchical clustering · ultrametric · prediction · negative result