Connecting to the bigger picture

Deep research agents are the multi-agent orchestration of Chapter 11 (MasterAgent, SubAgents, ReviewAgent), the memory mechanisms of Chapter 5, and the anti-hallucination contract of Chapters 10 and 13, all specialized for sourced output, plus a new training axis (SFT, RL/GRPO, CBR) that improves the model rather than the harness. The "harness matters more than the LLM" conclusion ties the whole guide together. The capstone borrows the review-pass idea as an optional fact-checking step.