In an exact model, reliability gave the full positive stage contrast range; two human MEG and fMRI tests failed

We tested visual stage matching in public human MEG and fMRI maps. The stage contrast compares early and late MEG with early visual and inferior temporal cortex. We removed image and broad class terms. A 92-image panel supplied development. A local code lock preceded new scores in separate 118-image and 78-image experiments, each with 15 adults per modality. Unknown person links were covered by bounds over all possible ten-person complements to five-person references. Both shared claims failed: positive stage matching and its reduction after one fitted product was removed. Holm p values were 1.0000. Mean full contrasts were 0.0286/0.0150 in the 118 panel and 0.0542/0.0621 in the 78 panel, for MEG/fMRI targets. Joint person, reference and image refit ranges crossed zero. An exact model with one identical true neural pattern and image-specific orthogonal measurement errors gave C=2(1-rho). It covers the full nonnegative contrast range [0,4] after the same class projection. The model is not fitted to people. It shows that positive matching alone cannot identify neural content change. An exploratory linear assignment tool gives exact point ranges when person links are absent. These ranges are not confidence intervals. A separate calculation checked all 360 person scores within 4.6e-15. Source window selection, small cohorts, source filters and conservative bounds limit scope. The package gives code, source pins and terms, every result, figures, plans and repeat steps. Original data remain external. No positive shared neural generalization or causal mechanism was confirmed.

human neuroscience · MEG · fMRI · visual stage matching · measurement reliability · representational similarity · negative result · unknown person links · reproducibility