In one mouse retina, image position reduced motion vector test error in a short trailing edge window but not a broad window

We tested whether image position predicts motion response vectors beyond cell type in matched calcium and electron microscopy data from one mouse retina. The main panel had 273 cells in 39 types outside eight known direction selective types. Each test excluded a whole recording. For the main broad response window, adding centered image coordinates increased squared vector test error by 0.44%. For a planned secondary window of three frames near the bar trailing edge, position reduced error by 17.76%. Seven of nine recording tests improved. A shuffle within type and recording gave a Monte Carlo p value of 0.0005, or 0.0055 after Holm correction for 11 initial cases. A later model used a fixed anatomical coordinate map and time slopes from separate response repeats. It reduced trailing edge error by 17.13% and 18.06% in the two repeat directions. A first order shift law explains how a direction-blind response can acquire a vector in a fixed window. These results support a possible timing contribution to short-window motion vectors. They do not establish its cause or explain the main broad-window bias. The package contains source code, derived data, all predictions and shuffles, figures, pinned provenance, independent checks, and source and offline repeat instructions. The scope is one retinal patch, one speed, fixed direction order, and the stated response measures.

retina · connectome · structure and function · calcium imaging · response timing · recording geometry · motion selectivity · reproducibility