rafaelsouza-tech/

rag-eval-ptbr

End-to-end RAG evaluation golden set in Brazilian Portuguese, tax-law domain (CTN — National Tax Code — and LC 123/2006, Simples Nacional). Designed to measure faithfulness/groundedness, context precision/recall, retrieval quality and citation accuracy of RAG pipelines — not a reading-comprehension QA set. Built for (and consumed by) the open-source harness rag-eval-ptbr, which adds a statistically grounded regression runner (rageval… See the full description on the dataset page: https://huggingface.co/datasets/rafaelsouza-tech/rag-eval-ptbr.