A fixed test did not confirm shared category block determinants in MEG and fMRI for 92 images in 15 adults

We tested a stricter form of shared visual image relations in human MEG and fMRI. A category signal with separate image gains can survive ordinary category removal. Within each class block, we removed row and column means and computed all two by two determinants. This transform is zero for an exact additive-plus-gain block. Development used 78 photographs from one study. A local code lock preceded a fixed test with 92 images and 15 released person rows per modality from another study. The same people supplied both modalities within each study, but their row links were unknown. Bounds covered every possible group of ten people outside five reference people. Early MEG and EVC gave possible mean score ranges of -0.00066 to 0.01091 and -0.00097 to 0.01665. Late MEG and IT gave -0.00179 to 0.01307 and -0.00388 to 0.00889. All outer 95% person intervals included zero. Both fixed shared tests failed; their two-claim Holm values were 1. Ordinary centered blocks retained positive mean bounds. The stricter fMRI maps showed weak agreement between people. All negative and discordant checks remain. A Cauchy-Binet implementation reproduced the same inner products with smaller arrays. An exact model shows that averaging category maps with changing image gains can create nonzero determinants. Thus even positive agreement would not identify neural source count or a cause. The contribution is a checked assay, a fixed negative benchmark, and reusable code. Source selection, fixed images, unknown person links, source processing, and missing independent trial estimates limit scope. Source arrays and photographs remain external. The package gives source pins and terms, plans, all results, figures, audits, and repeat steps.

human neuroscience · MEG · fMRI · representational similarity · category blocks · image reliability · determinants · negative result · Cauchy-Binet