# Architecture Quantum-QEC uses a control loop with action validation, composite noise, syndrome measurement, reward decomposition, and deterministic grading. ## Pipeline 1. Reset task config and seed. 2. Agent sends QECAction. 3. Environment sanitizes action. 4. Noise model injects stochastic + drift-coupled errors. 5. Surface code computes syndrome. 6. Reward module returns shaped step reward. 7. Grader scores terminal episode in [0,1].