One stored trial value could reverse the larger-response direction in 147 of 310 C. elegans pairs with unequal chemical counts

We tested the stability of a target used to compare wiring and measured neural responses. The target selects the direction with the larger absolute mean calcium response within a neuron pair. We used a pinned C. elegans WT atlas export. Each direction had at least five finite stored values. Of 310 pairs with unequal pooled chemical counts, one possible stored-value omission reversed the selected direction in 147 pairs, or 47.4 percent. We searched all possible single omissions. This search was added after the split results were known. Disjoint random halves agreed on direction in 51.5 percent of partitions. Independent half draws, which can share records, gave 64.4 percent agreement. A constructed example proves that the partition score is not a biological noise ceiling. Among 163 pairs that resisted every single omission, the count rule agreed with full response order in 75 pairs. The results identify target sensitivity. They do not show that stored values are wrong or that anatomy lacks functional information. Missing animal and event IDs prevent a test across independent biological samples. The package preserves new code, source hashes, pair scores, omission ranges, two figures, and repeat instructions. Direct scalar checks confirmed all 147 reversals. A fresh-source replay matched nine numerical files byte for byte.

connectome · Caenorhabditis elegans · signal propagation · response direction · target sensitivity · trial omission · reproducible benchmark