Adding a count of paths within one map reduced Brier error by 0.000014 in a fixed test of 21,074 C. elegans pairs without recorded direct edges

We tested whether path source maps add information about neural response detection. We used four anatomy exports and a pinned WT signal atlas. The graph retained 298 neurons. The main sample had 21,074 measured ordered pairs without recorded direct edges. Of 1,378 pairs with two-edge paths formed only by pooling, 42 had q < 0.05 (3.05 percent). The rate was 3.12 percent without a two-edge path and 5.28 percent with a path in one source map. Fixed logistic models controlled for neuron identity, trial count, soma distance, reverse chemical count, and degree. Ten folds kept opposite directions and bilateral copies together. Adding the number of relays supported by one map reduced Brier error from 0.03512860 to 0.03511418. The gain was 0.00001442, or 0.041 percent. A central 95 percent node weight range was -0.00004351 to 0.00007799. It includes zero and keeps fitted predictions fixed. The added count gave small losses with chemical edges only, two modern adult maps only, and an edge cutoff of two. Forty-seven relays with no response record supplied possible paths for 1,591 main pairs. The package preserves code, factual input extracts, source hashes, all results, figures, and repeat instructions. Independent route and arithmetic checks passed. A fresh source rebuild matched all 42 compared numerical files. This test withholds atlas entries. It does not establish signal routes or accuracy in new animals.

C. elegans · connectome · signal propagation · graph pooling · path provenance · electrical synapses · structure and function · prediction benchmark