We applied Karpathy Auto Research to chess through Opal, and the results were strong. Instead of relying on opening books or memorized theory, the system learns directly from outcomes. The agent plays matches against itself, evaluates positions with an engine, updates the policy, and runs the loop again. Over time, that feedback cycle compounds. Self-play -> Evaluation -> Policy update -> Repeat. The result was a +596 ELO improvement. When the learning loop is tight, improvement accelerates quickly.