A RAG system for a healthcare organization needs to answer clinical questions from medical literature. The system retrieves passages from 500,000 clinical papers. During evaluation, the team discovers that when the correct answer requires integrating information across three separate papers, the system either answers with the most statistically common finding or refuses due to insufficient context. What architectural addition directly addresses this multi-document synthesis gap?

The recursive retrieval and synthesis loop is correct because the failure mode — inability to integrate information across three papers — requires iterative, query-driven retrieval. A recursive retrieval loop uses Claude to identify what information is still missing after reading the initial retrieved context, then issues targeted sub-queries to retrieve the complementary papers. This mirrors how a human researcher would sequentially build up a complete picture from multiple sources. The option that increases the number of retrieved passages to 30 is wrong because adding more retrieved passages does not guarantee the specific combination of three papers needed is included — with 500,000 papers, statistical coverage is low, and a large retrieval count degrades answer quality by adding noise. The option that switches from dense vector retrieval to sparse BM25 retrieval is wrong because the problem is synthesis across multiple documents, not keyword matching precision — switching to BM25 changes the ranking mechanism but does not add multi-document synthesis capability. The option that pre-computes all possible two-paper and three-paper combinations during indexing is wrong because it is computationally infeasible at 500,000 papers, with three-paper combinations reaching the quadrillions, and it does not generalize to future papers.