Reinforcement learning with verifiable reward models cannot easily optimize for consumer product taste and user experience design