Stepping Out of the Shadows: Reinforcement Learning in Shadow Mode