With 18 candidates, uniform random choice among names absent from target lists gave 15.7% expected source accuracy in 415 C. elegans events

We found a source-identity cue in the target lists of a derived optogenetic response table. The fixed panel had 415 events, 18 candidate sources and 81 C. elegans worms. The original target lists omitted the actual source. A uniform random choice among candidate names absent from each list therefore had 15.68% expected accuracy without reading fluorescence. The per-event success probability is exactly 1/(K-m), where K is the candidate count and m is the number of candidates present as targets. This is a calculated selection effect, not measured neural decoding performance. The audit was added after the main fits. We also tested whether wiring scores help identify a source from other neurons' calcium responses. The controlled panel had 9,859 target responses. All candidate sources were removed from every response target list. Whole animals were kept apart during fitting and tuning. The reference decoder used observed target names and five canonical soma geometry matches. Inner tuning omitted all nine added wiring matches in all five outer folds. Thus the two final models had equal predictions. This equality follows from the selected rule and does not show biological equivalence. Geometry raised exact source accuracy from 5.59% to 12.23% and reduced log loss by 3.49% against the fitted observation-only control. Its fixed-prediction whole-worm range was 1.01% to 5.90%. However, geometry log loss was 2.8927, compared with 2.8904 for equal source probabilities. The result limits this decoder and its time-averaged fluorescence measure. It does not test all connectome models or identify a transmission mechanism. The package preserves code, exact inputs, raw-source hashes, all predictions, model states, controls, figures and repeat instructions.

connectome · source identification · calcium imaging · whole-animal validation · geometry · C. elegans