A log speed term gave no clear gain in 9A calcium prediction across eight flies after movement state and threshold control

We tested whether continuous leg speed improves calcium estimates for fly 9A axons after movement state and speed threshold control. These neurons belong to an inhibitory circuit onto hook movement sensors. We used Dryad version 5 from Dallmann and colleagues. Fixed rules retained eight flies, 96 trials, and 33,223 bins of 0.1 seconds. Each outer test excluded a whole fly. Inner tests selected ridge penalties using training flies only. Adding one filtered log speed term increased equal-fly mean squared error from 61.456 to 62.759 source z-score units squared, an increase of 2.12%. Resampling whole animals and repeating all model fits gave a 95% error-reduction range of -13.55% to +4.09%. Removing fly 8 from the final error average changed the reduction to +0.10%. A separate test across trials within each fly gave an uncertain 1.52% error reduction. Linear speed, two other calcium decay times, 20 Hz bins, and scoring inside active bouts gave no clear added gain. A source graph check confirmed 393 retained synapses from the MANC left front chief 9A to five hook axons. These counts set the anatomical context; they did not set fitted gains. The source study already reported a broad movement-state signal. This contribution supplies a compact benchmark, complete test predictions, and refit checks. Population calcium, source smoothing, eight flies, and a simple speed law limit inference. The results do not establish absence of speed coding or prove a binary inhibitory gate. Code, data provenance, derived features, results, figures, and repeat instructions are preserved.

connectome · presynaptic inhibition · Drosophila · proprioception · 9A neurons · calcium imaging · model transfer · cross-validation