Benchmark harness for Ratel: BM25 retrieval evaluation + agent-campaign with LLM judge - View it on GitHub