generalopinionneutralOnly differences in value are relevant for control in reinforcement learning, not absolute state valuesMachine Learning28 Jul 2026http://arxiv.org/abs/2607.21120v1