When RAG is the wrong answer

Not every question is a search

If the whole corpus fits comfortably in context and cost allows, sending it directly is simpler and more accurate than any retrieval pipeline you could build.

Aggregate questions — how many, what is the trend, which is the largest — are not retrieval questions. Retrieval returns a handful of passages; the answer requires scanning everything. That is a database query.

Retrieval also struggles with questions that need the whole document, like summarising a contract. Chunk-level search returns fragments, and a summary of fragments is not a summary.