A duration response rule did not pass a shared prediction test in five ECoG patients and six fMRI adults

We tested whether local single-image response profiles improve repeated-image predictions beyond a common profile. Three ECoG patients supplied a weighted ridge fit. A local lock preceded new scores in five other ECoG patients and a separate fMRI study. Mean normalized loss gain was +0.00471 in ECoG, with a 95% person t interval from -0.02500 to +0.03442. The fMRI release had eight records from six adults and unknown repeated-person links. Across all 72 links, mean gain was -0.01673 to -0.00013, with an interval union from -0.05682 to +0.04139. The fixed shared rule failed, with joint one-sided sign-flip p 0.96875. Negative and discordant controls remain. An added proof shows that this RMS-normalized affine prediction family need not be preserved under positive linear pooling. Two synthetic sites followed the rule exactly; pooling them gave prediction error. This is a conditional mathematical limit, not a measured BOLD pooling law or an explanation established by the human data. Clinical sampling, few training people, source selection, uncertain fMRI estimates and different cohorts and sensing scales limit scope. These are new endpoint tests in reused cohorts. No positive cross-modal neural or causal claim is made. The package gives compact inputs, source versions and terms, all outcomes, code, a prediction tool, plans, figures and repeat instructions.

human visual cortex · ECoG · fMRI · temporal adaptation · prediction · spatial pooling · normalization · negative result