Expected return of taking an action in a state under a policy, central to Q‑learning and DQN.