Separate ON and OFF contact directions gave no clear added direction match within type in 27 cells from one mouse retina

We tested whether separate ON and OFF contact directions predict their own motion responses better than swapped layer labels. We used 27 matched ON–OFF direction-selective ganglion cells from one e2198 mouse retina patch. The fixed primary score compared correct and swapped cosine agreement after type mean subtraction. Correct pairing gave mean direction error 20.53 degrees; swapped pairing gave 20.55 degrees. The centered cosine gain was 0.00946, with a descriptive cell resample interval of −0.00746 to 0.02821 and a one-sided shuffle p value of 0.194. Single-cell removal could change the gain sign. A secondary score that retained moment magnitudes had nominal p = 0.0385 and Holm p = 0.154 across four scores. It did not provide independent confirmation. Conditional repeat-group fits also gave no clear primary gain. The contribution is a paired layer test with an exact conditional variance formula, compact derived data, and a scoped negative result. Contacts include incidental appositions. Supplied types partly use contact direction. The small sample, shared circuits, and fitted calcium components prevent a claim that ON and OFF circuits are equivalent. We preserve source versions and hashes, new code, results, figures, and full and offline repeat instructions.

retina · connectome · structure and function · direction selectivity · starburst amacrine cell · ON and OFF pathways · paired comparison · permutation test · reproducible analysis