At gain 0.5, visual path steps gave no clear timing prediction gain for 76 mouse regions in a 2024 IBL export

We tested whether paths from two visual entry regions help predict when mouse brain regions represent a visual stimulus. The model kept all 213 regions and 21,654 positive edges in the Oh 2014 projection map. The main test used 76 regions in a March 2024 IBL timing export. The baseline used brain division and recording coverage. At a fixed path gain of 0.5, adding mean path steps reduced test squared error by 3.0 percent. A central 95 percent range under random region weights was -7.4 to 12.1 percent. Symmetric wiring gave almost the same main error. Replacing the planned 2025 coverage counts with the 2024 table changed the forward result to a 1.1 percent error increase. Its weight range also included zero gain. Including all 108 matched finite visual times gave a 1.2 percent error increase. The results do not establish useful timing prediction from this fixed rule. They show why source dates and full graph coverage matter in this benchmark. The package preserves code, figures, source hashes, all predictions, and a successful repeat of 19 numerical files. The timing summaries do not provide animal-level uncertainty.

mouse connectome · visual response timing · structure and function · weighted walks · source version · prediction benchmark