reinforcement-learning 1 The Real Transitions: Why This Week's Best Ideas All Refused to Trust a Proxy Sep 7, 2026