Four cells selected by input coverage gave no clear prediction gain over random or spatial selection in a ridge test across 79 C. elegans worms

Can wiring guide a small set of recordings that estimates several hidden cell responses? We tested a fixed input coverage rule in a C. elegans stimulation benchmark. Observed and hidden cell sets were disjoint. Left/right partners shared a set. The main panel had 3,554 animal-direction means from 468 events in 79 worms. Four selected downstream cells supplied green and red responses. Source early activity and hidden-target baseline activity remained available. Each test animal was absent from ridge fitting and penalty selection. The input rule raised mean anatomical coverage from 0.205 for random selection to 0.337. It gave 0.33% more prediction error than the mean of ten random rules. Its relative error decrease had a conditional 95% animal range from -2.11% to +1.47%. It gave 0.67% more error than spatial selection, with a range from -4.04% to +3.46%. Ranges keep predictions fixed and omit fit uncertainty. Budgets of one, two and eight cells also gave no clear selection gain. A second cell split exposed unstable static geometry predictions in one worm. A later check without those terms retained no clear gain over random selection. The result limits this rule in a retrospective selected panel. It does not establish equivalence or reject other recording designs. The package preserves a disjoint-set benchmark, source data versions and hashes, code, selected cells, model states, all results, three figures and repeat instructions.

connectome · C. elegans · structure and function · recording selection · shared input · calcium imaging · missing responses · animal validation