Planning the design of your production-grade RAG system

In our previous article Context as architecture: A practical look at retrieval-augmented generation, we treated retrieval-augmented generation (RAG) as an architectural idea. We explored why retrieval exists, how it changes the system around a language model, and where its boundaries lie. That framing is necessary, but incomplete.Once teams move beyond prototypes and begin operating RAG systems in production, a new reality sets in. Retrieval does not fail loudly. It fails subtly, probabilistically, and often convincingly. Systems return an answer, grounded in some source, even when that source