StaleBench v0.1.0

First public release of StaleBench, a benchmark that measures answer freshness in RAG systems.

1.It measures catch-up latency, which is how long after a fact changes the system gives the new answer, and the
2.share of answers that stay stale. It works with any RAG system through a simple two-method interface, so it is not tied to one retriever, vector store, or language model.

Included: the benchmark code, a reference RAG system, tests, and example results across several models, retrievers, and settings.