With source groups excluded, nine wiring terms raised error by 0.19% for the variance of three transformed values in 6,270 C. elegans response lists

We tested whether wiring helps predict the spread of stored calcium responses after neural stimulation. We used four native C. elegans chemical and electrical maps and the Randi wild-type atlas. All neuronal edges were kept, with 301 native labels available as relays. The panel had 6,270 ordered pairs with at least six finite response values. Each random partition used three asinh(x / 0.1) values for response controls and three other values for the sample variance to predict. Five folds excluded every training pair that touched a test source group. We fixed the models and averaged errors over 32 partitions. Nine wiring terms raised mean squared error from 0.380948 to 0.381661, an increase of 0.1872%. A descriptive central 95% endpoint-group weight range for error reduction was -0.6447% to 0.2777%. All eleven planned alternative tests gave a negative mean wiring gain. An exact diagnostic, added after the prediction results, gave a ratio of expected residual variance cross product to expected half power of 6.3060% after endpoint-group terms. This ratio describes random disjoint samples from fixed lists. It is not an animal variance fraction or a neural noise ceiling. The export has no event or animal IDs. The study therefore does not separate trial from animal variation or establish that anatomy has no effect on variability. The package preserves code, source hashes, benchmark inputs, predictions, exact sampling formulas, figures, and repeat steps. A fresh source rebuild matched all 84 result files.

connectome · Caenorhabditis elegans · response variance · signal propagation · fixed-list sampling · prediction · chemical synapses · electrical synapses