How do you eval RAG in production without a huge golden set?

We only have ~30 labeled queries. Curious what lightweight eval loops others run weekly before expanding coverage.