A tie for the best axon order required 20 input label changes in one mouse table with target and axon totals fixed

We tested the stability of a structural axon order in one P6 mouse omohyoid muscle source table. The table has six axons, 297 target rows, and 651 binary partner links. A target scored one when its inputs occupied consecutive ranks. All 360 orders were searched after removal of reversal. One input label change removes a present axon and adds an absent axon in a target. With every target input count and all six axon totals fixed, 19 changes could not end the unique best order. Twenty changes could give a tie, and 21 could give a strict loss. When axon totals were free, ten changes sufficed. Exact first loss of forced neighbor pairs required 20 to 36 changes. A planned check on 52 targets with three or four inputs preferred another order, with score 43 instead of 37 for the full table order. Thus, the full score preference came from targets with two inputs. We provide exact lower bound certificates, integer examples, original code, source provenance, aggregate facts, and figures. A separate checker verified 652,662 inequalities and 1,795 integer examples. These limits concern the stated score and error models. They do not estimate reconstruction error or measure motor recruitment.

connectome · motor recruitment · axon order · annotation errors · consecutive ones · exact verification