Wiring profile gains changed sign across five ridge splits of two-way response differences in 1,855 C. elegans atlas pairs

We tested whether complete wiring profiles predict differences between opposite stimulation directions in a fixed C. elegans calcium atlas. We used all 1,855 exact neuron pairs with at least six stored response values in each direction. Graph calculations retained all 302 canonical neuronal slots in five anatomical maps. The main ridge baseline controlled neuron effects, measurement counts, direct and path counts, and soma geometry. Adding 120 antisymmetric profile products raised mean squared error by 1.52% in the fixed main split. Four other planned splits gave decreases of 0.50% to 1.25%. For reciprocal means, symmetric profiles reduced error by 1.26% to 5.66% across all five splits. Both main endpoint-weight ranges included zero. A second test reproduced an earlier model on 6,270 response directions exactly. A symmetric profile term retained 92.37% of its full 4.62% error decrease. Thus, the earlier aggregate profile gain did not establish a stable gain for direction differences. This is a prediction test within one summary atlas. It does not test biological reciprocity or prediction in independent animals. Animal and event identifiers are absent. The package includes complete graph arrays, the paired benchmark, code, source hashes, all predictions and model states, figures, and repeat instructions. A source rebuild and complete numerical replay passed.

connectome · C. elegans · signal propagation · response direction · pairwise prediction · symmetric kernels · reproducibility