A fixed test did not confirm image isolation across photographs in MEG and fMRI from 15 adults

We tested whether within-class image isolation repeats across different photographs. Public MEG and fMRI targets came from one cohort of 15 adults and two photographs for each of 78 matched descriptions. Isolation is the mean distance of an image from the other images in its class. We averaged correlations across five classes. A local plan and code lock preceded new scores. Missing person identifiers led us to allow every bijection between photo-set rows, separately within each modality. Assignment optimization gave exact mean and positive-count bounds over 15! links. Late MEG possible means ranged from -0.05974 to 0.22907; IT fMRI means ranged from -0.17822 to 0.16997. Only two positive MEG scores and zero positive fMRI scores were guaranteed. The fixed shared sign rule failed, p=1.0. A MEG image-label reference was favorable, but its wider person and description range crossed zero. fMRI checks remained uncertain. An exact model gave perfect isolation transfer from repeatable error sizes despite identical true image responses. Independent scalar scores, a separate dynamic program and a clean source-based replay checked the results. The contribution is a negative benchmark and a tool for conservative median inference with missing links. It does not confirm cross-modal neural generalization or absent invariance. One cohort, source choices, uncertain image order, measurement filters and noise limit scope. Code, source hashes and terms, all outcomes, figures, proofs and repeat steps are supplied. Original neural arrays remain external.

human vision · MEG · fMRI · image isolation · missing person links · assignment · negative results · representational geometry