Two-step chemical routes gave no clear response prediction gain in 13987 worm pairs without direct chemical edges

We tested whether two-step chemical routes predict evoked response detection in worm neuron pairs with no direct chemical edge in either of two adult maps. We linked the Witvliet D7 and D8 maps to the Randi wild-type atlas by cell labels. There were 13,987 directed pairs and 721 detected responses. Models controlled endpoint degree, electrical count, reverse chemical count, observation count and cell identity. Five folds held out source cell classes, with left and right partners kept together. Adding mean route count reduced log loss from 0.185027 to 0.184258. The reduction was 0.000769, with a conditional 95% crossed-class interval from -0.001205 to 0.002602. All six sensitivity intervals included zero. A synthetic route effect gave a positive gain. This result does not establish that indirect routes have no effect. Separate anatomy and activity animals, incomplete wiring coverage, missing spatial controls, pooled responses and one prediction model limit the test. Code, pair evidence, uncertainty draws, hashes and a figure support reproduction.

C. elegans · indirect paths · evoked responses · connectome · held-out cells · null result