Forward treadmill speed gave an uncertain 5% calcium prediction error decrease beyond one joint's motion in a ridge test of eight flies

We tested whether forward treadmill velocity adds information to front leg joint motion in fly 9A axon calcium. Published connectomes place these inhibitory cells between descending input and movement-sensing axons. We used 96 recordings from eight flies in Dryad version 5. A reference ridge model used joint angle, velocity, nonlinear terms, and causal histories. Each test excluded one whole recording within a fly. Adding treadmill velocity gave a mean relative error decrease of 4.99%. The 95% Student mean interval was -3.95% to +13.93%; five of eight flies had lower error. A whole-animal percentile bootstrap gave +0.03% to +12.87%, so the mean inference depends on the interval rule. Extra histories, future joint motion, three treadmill channels, and two other scale floors also gave Student intervals that included increased error. The first estimator had extreme predictions from almost constant angle-history features. After inspecting those fits, we imposed a feature scale floor of 0.01 and retained all initial results. This amendment was not registered in advance. A source-graph audit found that descending cells supplied 33.1% to 51.6% of retained chief 9A input contacts. The imaged population is not a matched chief cell. The study provides a conditional function benchmark and a documented scaling failure. It does not identify a causal motor signal or measure inhibition. The package contains code, derived inputs, all predictions, model states, source hashes, figures, a scaling bound, and exact repeat checks.

connectome · Drosophila · proprioception · 9A neurons · presynaptic inhibition · calcium imaging · treadmill velocity · joint motion · ridge regression · feature scaling · prediction