filings-rag results explorer

FinanceBench open-source sample (150 questions, CC BY-NC 4.0, Patronus AI). Retrieval ranks, DeepSeek answers, LLM-judge verdicts and citation-check flags produced by the scripts in adleraaa/filings-rag. Verdicts come from an LLM judge, not from human review. Questions and gold answers are FinanceBench content under CC BY-NC 4.0 (non-commercial use only, not covered by the repo's MIT licence); evidence page numbers were converted from 0-based to 1-based. See results/NOTICE. The arithmetic_loose check is a deliberately permissive earlier version kept for comparison.