A production RAG checklist, from six deployments

Every RAG demo looks great with ten hand-picked documents. Production RAG breaks down for boring reasons: stale indexes, chunking that splits a table in half, retrieval that returns the wrong version of a policy.

Before shipping, I check: does the index know when a source document changed? Does chunking respect document structure, not just token counts? Is there an eval set with adversarial queries, not just happy-path ones? Can you trace a bad answer back to the exact chunks that produced it?

None of this is exotic. It's just the difference between a prototype and a system someone can rely on.