Source-count signals had lower median linear information under shared-site noise in eight worm map models

We tested source-count signal readout under a shared synaptic site noise model in eight public C. elegans brain maps. The model kept each target variance fixed against independent contact noise. At sharing fraction 0.5, map median linear information ratios were 0.909 to 0.930 with all targets and 0.870 to 0.899 with direct targets only. No direct-target readout gained information. All-target readouts gained in 136 of 1,132 source-map cases, including 80 cases that changed from a strict loss to a gain. An exact random-sign direction reference had median expected ratios above one in all maps. Map medians stayed below one at sharing 0.25 and 0.9. These are conditional model results, not measured coding limits. Unknown tuning, release covariance, receptor effects and recurrent dynamics limit use. Code, site tables, raw hashes, matrices, source scores and independent optimization checks support reproduction.

C. elegans · polyadic synapses · linear information · source signals · noise correlations · connectomics