An order learned from 148 targets gave 80% test agreement beyond count controls in one six-axon mouse table

We tested whether an axon order learned from some muscle targets also describes inputs to separate targets. We used the Figure 5a source table of Meirovitch and colleagues. It has 297 target rows and six axon columns from one developing mouse muscle. We chose an order on 148 rows by testing all 360 choices up to reversal. Inputs at 119 of 149 test targets occupied consecutive ranks, or 79.87%. Two graph-control chains kept every target input count and each axon's target count separately in each half. Their 1,998 retained graphs gave mean test agreement of 34.90%. None reached the observed score. The Monte Carlo upper-tail estimate was 0.0005003. Mean agreement across 1,000 further splits was 79.03%. A label audit matched all 15 pair counts in the printed figure after reversing the CSV column labels. The result shows a repeatable order beyond connection counts in this table. It does not measure recruitment, synchrony, or the cause of synapse removal. The reference does not keep geometry. This is an exploratory test of one muscle and a table that omits single-input targets. The package has independent code, aggregate results, figures, source hashes, sampler checks, and repeat instructions. Raw source files remain external under their original terms.

neuromuscular junction · connectome · axon order · synapse removal · fixed connection counts · separate target test