With independent white noise and free LM drive, averages at 16 S outputs kept at most 10.23% of uncoupled information in one passive human foveal model

We tested whether voltage averaging preserves an S signal when LM drive is unknown. We used one archived human foveal contact graph with 333 units and 1,141 rows. The passive model had equal leaks, equal capacitances, independent white source noise and stationary constant signals. We compared exact Gaussian information at the same duration with coupling set to zero. With a free constant LM amplitude, the 16 S outputs retained 10.1137% of reference information at ten membrane time units. Their long-window ratio was 10.2237%. A covariance bound made this an upper bound for every averaging window. With LM amplitude fixed, the same panel retained 97.8989% at ten units. Adding the 56 direct LM neighbours raised the free-LM ratio to 98.4881% and gave full asymptotic recovery. The known static boundary panel therefore also rejected the LM nuisance in the long-time white-noise model. Four label assignments, altered coupling, rod-edge removal and equal pair weights retained the panel difference. Added observation noise limited full recovery. A later finite-prior test showed a smaller loss when LM amplitude was tightly known. The free-nuisance result has a singular zero-coupling limit and must not be read as a general visual deficit. Independent source readers, lag-covariance ODEs, Gaussian calculations and 37 checks passed. A clean repeat matched 12 numerical files. The package preserves source data, provenance, code, proofs, results, figures and repeat steps. These are fixed-model results, not measurements of human colour vision or a test of all temporal decoders.

human fovea · electrical connectome · S cones · LM nuisance · temporal averaging · Gaussian information · observation design · source noise