`run_episode()` would simply return the most recent reward. This makes more sense imo.