Under a binary source sign model, unknown signs left the covariance of inhibitory input fraction with model motion selectivity undetermined in 8,061 MICrONS neurons

We tested recorded inhibitory inputs against model motion direction selectivity in one MICrONS v1507 mouse volume. A full scan of 303,164,967 connection records gave 27,308,102 incoming synapses for 8,061 selected targets. Consistently typed sources covered 8.66% of these synapses. Their inhibitory fraction and type shares reduced prediction error by 0.105%, with a conditional 95% block range of -0.177% to +0.400%. This gave no clear added gain. We then kept all broad source signs fixed and gave each unknown source segment one binary sign for all its targets. Broad signs covered 8.84% of inputs. Under this model, exact bounds on covariance of total recorded inhibitory synapse fraction with the baseline residual of model selectivity were -0.014701 to +0.014730. Complete source sign assignments reached both endpoints. Both signs remained possible when all unknown segments that reached only one selected target had to share one sign. Median per-target inhibitory fraction bounds were 6.27% and 97.36%. A secondary spatial score gave small gains that varied across splits. The bounds identify missing information, not real transmitter labels. Source fragments, count-based strength, model labels and one mouse limit biological claims. The package preserves code, source hashes, selected summaries, predictions, exact bound witnesses, figures and repeat steps.

connectome · MICrONS · inhibition · direction selectivity · source coverage · partial identification · synapse counts