Benchmark Methodology
How we measure Qefro’s accuracy and refusal behavior.
Methodology
We evaluate Qefro on a fixed set of question–answer pairs drawn from customer-style knowledge bases (policies, product docs, FAQs). Each query is scored as correct, appropriate refusal (no relevant source), or incorrect (hallucination or wrong citation). Scores are computed per category and release.
Test set composition
Benchmarks include factual lookups, multi-step policy questions, out-of-scope queries, and ambiguous phrasing across English and multilingual samples. Knowledge bases range from small FAQ sets to larger document collections so results reflect real deployment sizes.
Results
Published accuracy and refusal metrics are updated when we ship meaningful RAG or model changes. Contact support@qefro.com for the latest benchmark report for your industry or use case.
Limitations
Benchmarks measure retrieval and answering behavior on curated test sets; they do not guarantee performance on every production corpus. Your content quality, chunking, and access rules materially affect live accuracy.