Run the same question with RAG and with a direct LLM (full document text). See answers and token usage side by side.