Proving RAG Systems: Two Deterministic Benchmarks for Regulatory Compliance

As regulatory requirements for AI systems tighten—particularly the EU AI Act's audit obligations taking effect August 2026—most RAG systems lack the transparency to prove their decisions meet compliance standards. These pre-registered benchmarks fill that gap by measuring both auditability (can you replay a decision?) and temporal accuracy (was the information used correct when accessed?), shifting RAG evaluation from 'how good are the answers?' to 'can you prove how you got them?'