A neuron removal bound exceeded exact target cuts for 8,336 pairs in eight C. elegans maps under whole site loss

One presynaptic site can contact several neurons. We tested how this shared structure changes cuts at a target when removal acts on the whole site. A prior SCI audit removed the target neuron before counting its reachable input sites. That count is feasible but need not be minimal. We prove an exact rule: remove all sites that contact the target, then count those whose presynaptic neurons remain reachable from the source. Each counted site is forced, and the set is sufficient. In eight fixed C. elegans chemical graphs with 162 neurons, the old target count was too high for 8,336 of 164,226 reachable ordered pairs. The correction removed 1,225 cases where a global cut was placed below both single endpoint optima. Published global cuts remain valid. We then tested a corrected cut deficit in 11,504 measured WT optogenetic atlas pairs without direct chemical edges in either adult map. Nested logistic models controlled for neuron identity, measurement coverage, distance and wiring. Adding the deficit reduced Brier error by 0.124%. A descriptive neuron weight range for fixed predictions included both signs. Scope checks gave small gains and losses. The exact correction does not establish functional resilience. Independent flow, exhaustive deletion, source rebuild and replay checks passed. The package preserves the proof, code, small factual inputs, all corrections, predictions, figures, source hashes and reproduction steps.

connectome · C. elegans · polyadic synapse · minimum cut · signal propagation · structure and function