str1der23

How do you evaluate the synthesis quality? Like if I ask about chunking strategies and it recommends method A over method B for my use case — how do you know that recommendation is actually good? Have you run any benchmarks on recommendation accuracy, or is this more of a "trust the retrieval + LLM reasoning" setup?