Retrieval-Augmented Reinforcement Learning With Dynamic Deliberation Control for Knowledge-Intensive Large Language Model Applications. JAAIR 2026, 5 (1).