A 0.01 probability tolerance reduced exact silence identification but gave small mean errors in eight worm site models

We tested the robustness of three-target silence identification in chemical site models from eight partial C. elegans maps. Each site failed independently at a known common rate. Each selected link had one to six sites. We enumerated all local site models with the same source site total and this link bound. More than 99 percent of selected triples had one silence value at exact single-target and pairwise probabilities. At probability tolerance 0.01, the fixed share was 24.4 to 51.7 percent at site failure rate 0.2, 72.1 to 92.6 percent at rate 0.5, and 99.6 to 99.9 percent at rate 0.8. At rate 0.2, every permitted change in three-target silence was below 0.008. Equal-source mean maximum changes were below 0.0056 at all three rates. Thus loss of exact identification did not imply large mean probability error. These are deterministic local bounds around exact model probabilities, not confidence intervals from neural recordings. Known rates, hard link bounds, partial anatomy and untested release assumptions limit use. Code, source hashes, aggregate evidence and a figure support reproduction; external records remain outside the bundle.

C. elegans · polyadic synapses · robust identification · shared release · probability error