Original data · currency across the catalog

Currency benchmark

How current every verification pack is against the moving frontier of agent capability — decay from the dated cohort it was last validated against, not calendar age. A stale pack is capped out of a top grade until it's revalidated.

Every pack · currency against the current frontier cohort

Pack	Method	Last validated	Currency	Lifecycle	Grade
Medical RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	B
Legal contract RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	B
Financial reporting RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	B
Insurance policy RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	B
Public-sector benefits RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	B
HR policy RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	A
Security-policy RAG — groundedness & abstention	RAG groundedness	2026-H1	1.00	fresh	A
Browser agent — prompt-injection red-team	Prompt-injection defense	2026-H1	1.00	fresh	B
EU AI Act — high-risk conformance	AI Act obligation checklist	2026-H1	1.00	fresh	A
General RAG groundedness — draft submission	RAG groundedness	2026-H1	1.00	fresh	F
Customer-support RAG — groundedness & abstention	RAG groundedness	2025-H2	0.66	aging	A
E-commerce policy RAG — groundedness & abstention	RAG groundedness	2025-H2	0.66	aging	A
API-docs RAG — groundedness & abstention	RAG groundedness	2025-H2	0.66	aging	A
Tool-calling correctness	Tool-calling correctness	2025-H2	0.66	aging	B
General-knowledge RAG — groundedness & abstention	RAG groundedness	2025-H1	0.32	stale	B

Currency decays from the dated cohort a pack was last validated against toward the current frontier snapshot (2026-H1). stale caps a pack out of an A until it's revalidated; expired pulls it from sale. Not calendar age — an eval the frontier has saturated no longer discriminates, however well-built.

Currency is one of the six authority axes — robustness across time, the sibling of the robustness axis's robustness across the input space.

← Back to benchmarks