Thanks Rui and Friedemann. The model in Nejad et. al. (2025) also contained a feedforward connection between L2/3 and L5, which muddles the mapping to a JEPA a fair bit imo