In hemibrain v1.2, edges below five synapses bypassed 1,420 of 1,423 two-step MBON–DAN cuts and kept median input-normalized weight of 89.8%

We tested whether a sole relay selected by a synapse cutoff carried most of a short anatomical feedback score. We used the hemibrain v1.2 export of 21,739 non-cropped traced neurons. Sources were 44 right-side mushroom body output neurons (MBONs). Targets were 157 right-side dopamine neurons (DANs). We excluded every pair with any recorded direct edge. At five synapses per edge, 1,423 pairs had exactly one two-step relay. We removed that relay separately for each pair. With all recorded edges restored, 1,420 pairs had a two-step bypass. The median retained input-normalized weight was 89.8%, with a middle 50% range of 66.8% to 97.8%. It exceeded half in 1,206 pairs. Output-normalized weights gave a median of 88.0%; count products gave 62.7%. A bypass with at least three synapses on each edge remained in 1,048 pairs. Removing all one-synapse edges lowered median surviving weight to 29.9% of the original full score. With a separate external-relay rule, 1,285 selected pairs kept median input weight of 73.6%. Sparse matrix products after explicit relay deletion matched all primary results. We supply the sufficient data extract, source hash, code, figures, exact neuron IDs, and bypass witnesses. These are unsigned two-step scores in one released graph. They do not establish functional feedback or rescue after neuron loss. Synaptic signs, compartments, missing cells, and false detected contacts can change the biological interpretation.

hemibrain · mushroom body · dopamine · connectome · two-step paths · synapse cutoff · relay removal · reproducible analysis