This line of work asks how representations evolve across recurrent processing in deep vision models. In a first thread (VSS 2025 / CCN 2025), we showed that default long-range modulatory feedback connections compacts category clusters and pulls exemplars toward their prototypes: refining local representational geometry while largely preserving global structure, an emergent “prototype effect.” A follow-up thread (VSS 2026 and an 8-page CCN 2026 proceeding) extends this to four different recurrent model families and shows that different recurrent architectures exhibit distinct geometric trajectories and decision-stage structures, inviting more careful differentiation among recurrent vision models in both computational and neuroscientific contexts.