Most RAG writing focuses on chunking, embeddings, and reranking — the “retrieval” half of retrieval-augmented generation. Far less is…