Pooling independent site noise removes zero-noise target combinations in eight selected worm map models

We tested exact covariance rank in shared-release models of eight selected C. elegans synaptic site maps. Separate source matrices had 137 to 251 null dimensions per map in total. Between 20 and 44 dimensions exceeded the loss forced by the incidence zero pattern. When all retained sites contributed independent positive-variance noise, the pooled matrices had full rank on all 140 to 162 retained targets. Integer elimination modulo a prime certified each pooled rank. Under equal site variances, minimum pooled correlation eigenvalues ranged from 0.122 to 0.386. We also prove that a zero-noise readout of one source cancels every mean signal produced by changes to release probabilities at those same sites. Thus a low source-level noise dimension does not alone show reliable representation. Fixed unit effects, independent release, selected cells and no paired activity measurements limit the result. Code, aggregate evidence, source hashes, a figure and external retrieval instructions support reproduction.

connectome · shared release · covariance rank · population coding · C. elegans · noise model