Half-cell recordings retained about 49 percent of mean source-pair information in two passive worm models

We tested how sensor count limits source-pair information in two passive adult C. elegans models. Equal independent Gaussian noise entered all cell currents. Sensor voltages had no added noise. A projection argument bounds mean retained squared separation over all input pairs between (k-1)/(n-1) and k/(n-1), capped at one, for k sensors and n cells. The bound holds for any invertible linear mixing. In two partial electrical maps with 147 and 149 cells, 40 random half-cell panels at contact weights and normalized leak 0.1 retained mean fractions 0.49330 and 0.49342. Only 10.9 and 12.3 percent of pairs retained at least 90 percent. All 96 tested map settings obeyed the bound. Direct covariance solves and 100 random small models checked the calculation. This is a squared Gaussian separation result, not retained mutual information or measured neural decoding. Added readout noise can break the lower bound. Partial anatomy, assumed conductances, exact covariance and no activity measurements limit biological inference. Original code, proof, aggregate results, a figure, source hashes and external regeneration steps are supplied.

electrical connectome · sensor count · source discrimination · input noise · partial observation · Gaussian information · C. elegans