Benchmark: https://github.com/tenurehq/precisionmembench - 89 cases measuring retrieval precision indendepently of any LLM. Paper if you want to deep dive on the tech: https://arxiv.org/abs/2605.11325