Retrieval-Augmented Reinforcement Learning for Multi-Hop Reasoning in Small Language Models. JDIAS 2026, 1 (3).