Four unsigned chemical direction terms gave no clear added lag prediction gain in 21 freely moving C. elegans recordings

We tested whether chemical wiring direction predicts the order of calcium activity in freely moving C. elegans. We used 21 baseline recordings, 154 named cells and 75,602 animal-pair records. Structural features kept all 302 slots of the corrected Cook reference graph. The target was the mean forward-minus-reverse correlation at 0.5, 1 and 2 seconds after specified behavior removal. A baseline controlled exact cell identity, zero-lag correlation, cell autocorrelation and graph degree. Each test excluded one complete animal. Inner animal groups selected separate penalties for the baseline and added wiring terms. Four unsigned direct-direction terms reduced mean squared error by 0.00988%. The conditional 95% animal resample range was -0.00084% to +0.01911%. Full-graph walks gave 0.00519%. Excluding both animals and the same named pairs gave a 0.00738% error increase. An exploratory follow-up found repeatable lag information: shrunken pair means from other animals reduced error by 4.93% against a constant mean and by 1.91% against the cell and timescale baseline. Matched half-recording lag correlation was 0.303 within animals and 0.068 between animals. These results limit the tested unsigned count features, not the biological role of chemical direction. Calcium filtering, scanning, unknown synaptic signs, different anatomy animals and behavior removal limit interpretation. The package preserves code, source hashes, all summaries, predictions, figures and repeat instructions. A full source replay reproduced all checked numerical results.

connectome · Caenorhabditis elegans · calcium imaging · lagged correlation · chemical synapse direction · animal transfer · nested validation · reproducible benchmark