Three of 21 mouse cortex groups had excess clustering of target categories under contact shuffles

We tested whether contacts on one axon share target categories more often than expected from pooled category totals. We used two public electron microscopy workbooks from developing mouse barrel cortex. We removed one seed contact, combined AIS seed subtypes, and retained 564 axon rows with consistent counts and at least two other contacts. We shuffled target labels while preserving each axon contact count and each group category total. Three of 21 layer, dataset, and seed groups passed a Holm test at 0.05 with 99,999 shuffles. The excess same-category probabilities were 0.0835 and 0.0910 for layer 2/3 AIS groups at P14 and P28, and 0.1677 for one layer 4 AD group at P14. The P28 excess fell to 0.0015 after a ten-contact filter removed all four retained axon rows with source subtype label 2.1. This filter changes subtype composition. We found 46 rows whose category sum differs from the reported total. We supply sensitivity results, exact null moments, checked code, derived scores, and source hashes. The result rejects a uniform contact allocation in selected finite tables. It does not establish an age effect across animals, physiological inhibition, or a rule for neural activity.

connectomics · mouse barrel cortex · inhibitory axons · target categories · sampling · conditional randomization