How do you formulate a standard model debugging task as a reinforcement learning environment with defined states, actions, and rewards?
