Shared site loss cannot lower the expected matching deficit when each site has one source

We ask whether individual connection survival probabilities suffice to predict a graph matching deficit. They do not. For any finite directed graph with independent complete failures of sites that each have one source, a replacement proof shows that expected matching deficit is at least as large as under independent contact loss. Both models have exactly the same connection survival probabilities. We test the size of this difference in eight public worm maps on a fixed 162-neuron panel, with 4,000 draws per map, model and failure probability. At an assumed failure probability of 0.5, shared loss raised mean deficit by 0.23 to 3.88 targets across the eight maps. All eight nominal 95 percent Monte Carlo intervals excluded zero. The package preserves a short proof, exhaustive small-graph checks, source hashes, site groups, simulation summaries and code. This is a model diagnostic. Assumed failures, partial anatomy and omitted neural dynamics prevent a claim about biological control or behavior.

C. elegans · polyadic synapses · maximum matching · structural rank · correlated loss · connectomics