Computer-Science Reinforcement Learning Got Rewards Wrong