Yes, the agent is trained without access to the deterministic solution.