Fine-tuning versus retrieval: the real performance numbers on claims data.

Everyone has opinions. We ran the benchmarks. Fine-tuned models outperform RAG on structured claims but lose on edge-case policy interpretation. The nuances matter.