A shared heat response variance rise was not confirmed in 115 EEG and 14 fMRI adults

We tested whether stronger heat raises trial response variance along a reader trained only on physical input. Public EEG and fMRI studies supplied separate healthy-adult cohorts. Development used 39 EEG adults and eight MRI adults. A local code and model lock preceded reserved variance scores. The fixed test retained 115 EEG and 14 MRI adults. Mean corrected log high/low variance changes were -0.0918 [-0.2089, 0.0253] and +0.4586 [0.0321, 0.8852], with brackets for 95% person t intervals under the stated model. Geometric ratios were 0.912 and 1.582. Changes relative to 64 fixed perpendicular readers were -0.0357 [-0.1519, 0.0805] and +0.3658 [-0.0824, 0.8141]. Both shared Holm P values were 1. Mean input reading was positive in each modality, but the variance-rise claim failed. We give a log-variance assay and its scope proof. Relative log changes cancel common trial sampling bias under a common-shape model; separable Gaussian covariance is sufficient. Direction-specific trial correlation breaks this condition. At the source designs, Gaussian models with equal marginal variance gave expected apparent ratios of 2.90 and 3.38. These are model examples, not fitted human causes. All 999-draw simulation cases, negative people, controls and exclusions remain. A fresh full-file audit rebuilt both MRI grids within 3.6e-15 source units, and an isolated replay reproduced every fixed result. The package gives permitted inputs, source versions and terms, code, plans, figures and repeat steps. Selected trials, source processing, sparse residual degrees of freedom, different cohorts and unequal signal filters limit scope. No new shared biological variance rise, common neural cause or clinical pain marker was established.

human neuroscience · EEG · fMRI · pain · trial variance · cross-modal test · temporal dependence · reproducible benchmark