Under slow independent noise and no added readout noise, 72 cone outputs were necessary and sufficient for full S-only detection in one archived passive human foveal model

We tested the smallest set of cone voltages that preserves S-only signal detection in one archived human foveal graph. The static passive model kept 333 named units and all 1,141 contact rows. It used slow independent Gaussian input noise and no added readout noise. For every positive finite coupling, the unique optimal voltage weights had support on 16 labelled S cones and their 56 direct LM neighbors. A proof showed that all 72 outputs were necessary and sufficient for the full one-sample squared detection score. At the primary coupling-to-leak ratio of 0.0545456 per weight unit, the S outputs alone retained 99.2662% of that score. This reproduces an earlier SCI result. A greedy 48-output panel retained 99.9023%; its size is feasible, not a proven minimum. With a common input-noise fraction of 0.01, the 72-output panel retained 94.8862% of the full-network score. For any positive common fraction below one, all 333 outputs were required for exact recovery under the tested common-noise model. Assigning two uncertain labels to S changed the independent-noise exact panel size to 76, 78 or 81. Independent source readers, current equations, output-deletion identities, synthetic graphs and a fresh run checked the calculations. The contribution is an exact observation rule and its test in one released graph. It does not establish a biological readout or human visual performance. The package preserves the source table, code, proof, provenance, all results, figures and offline repeat instructions.

connectome · electrical coupling · cone photoreceptor · sensory coding · noise covariance · observation design