Closes the simulation-to-reality loop by comparing predicted state with actual post-execution observations and routing model error into repair and learning.
