Large-scale neuronal circuit mapping using electron microscopy demands laborious proofreading by humans who resolve local ambiguities with larger contextual cues or by reconciling multiple indepen- dent reconstructions. We developed a new method that empowers expert neuroanatomists to apply quantitative arbor and network context to proofread and reconstruct neurons and circuits. We implemented our method in the web application CATMAID, supporting a group of collaborators to concurrently reconstruct neurons in the same circuit. We measured the neuroanatomical underpinnings of circuit connectivity in Drosophila neurons. We found that across life stages and cell types, synaptic inputs were preferentially located on spine-like microtubule-free branches, "twigs", while synaptic outputs were typically on microtubule-containing "backbone". The differential size and tortuosity of small twigs and rigid backbones was reflected in reconstruction errors, with nearly all errors being omission or truncation of twigs. The combination of redundant twig connectivity and low backbone error rates al- lows robust mapping of Drosophila circuits without time-consuming independent reconstructions. As a demonstration, we mapped a large sensorimotor circuit in the larva. We found anatomical pathways for proprioceptive feedback into motor circuits and applied novel methods of representing neuroanatomical compartments to describe their detailed structure. Our work suggests avenues for incorporating neuroanatomy into machine-learning approaches to connectomics and reveals the largely unknown circuitry of larval locomotion.