At fixed target scales and gain 0.8, source variances bounded by one million fitted 11 wing muscle correlations in two signed fly connectome models

We tested whether unequal independent source drive can remove a mismatch between fly wiring and wing muscle coactivity. We used every supplied connection in BANC v888 and MANC v1.2.1: 192,719 graph units and 18,926,219 directed pairs across two animals. The functional target was a mean correlation matrix of 11 muscles from 72,219 wingbeats on 31 recording dates. We fixed input-normalized count weights, predicted transmitter signs, target scales and path gain 0.8. We then fitted nonnegative source variances under a common upper bound Q, relative to equal unit drive. At Q = 10,000, minimum relative matrix error was 6.67 to 8.38 percent across the four sides. At Q = one million, all four targets fitted to numerical precision. The feasible relative error bounds, including estimated path error, were below 7.1e-7. We derive and verify lower bounds for the bounded covariance problem. Each primary fit had 66 strict-interior response generators of rank 66. A proof and explicit alternative drives show that its source variances are not unique. In an added test, 8 of 99 changed muscle-label matrices also fitted BANC, and 19 fitted MANC. An early-date fit gave the same later-date error as the early empirical mean. The result concerns a static incoming-drive model with fixed target scales. It does not measure true source drive or explain fast flight control. The package preserves complete primary response matrices, exact trace extracts, code, fitted drives, bounds, figures, source hashes and reproduction instructions.

connectome · Drosophila · motor control · source variance · covariance · identifiability · BANC · MANC · wing muscle