# AgentGrading.ai > A verification authority for AI-agent evaluations. We certify that an eval > can be trusted — reproducible, ungameable, and anchored to real external > standards — before it grades anything. Then we sell the eval-packs that > pass. We grade the eval, not the agent. ## Key pages - [Homepage](https://agentgrading.ai/): the authority thesis and the six verification axes (structural validity, discriminating power, standard coverage, thoroughness, robustness, and currency). - [Packs catalog](https://agentgrading.ai/packs): verified eval-packs, browsable by type — [capability](https://agentgrading.ai/packs/capability), [safety](https://agentgrading.ai/packs/safety), [conformance](https://agentgrading.ai/packs/conformance) — or by test method — [RAG groundedness](https://agentgrading.ai/packs/rag-groundedness), [tool-calling correctness](https://agentgrading.ai/packs/tool-calling), [prompt-injection defense](https://agentgrading.ai/packs/browser-injection), [AI Act obligation checklist](https://agentgrading.ai/packs/ai-act-checklist). - [Guides](https://agentgrading.ai/guides): methodology guides on how to evaluate AI agents (groundedness, tool-calling correctness, safety red-teaming). - [Benchmarks](https://agentgrading.ai/benchmarks): original discriminating- power data — how our packs score reference agents of known quality (good, broken, sabotaged). - [Standards](https://agentgrading.ai/standards): the external standards our conformance packs are anchored to (EU AI Act, OWASP-agentic, NIST AI RMF, ISO/IEC 42001). ## Notes for AI crawlers - All pack, guide, benchmark, and standard pages are statically generated and safe to crawl and cite. - Benchmark pages carry `Dataset` schema.org markup; pack pages carry `Product`; guides carry `TechArticle` + `FAQPage`, and timely analysis of a news event carries `NewsArticle` + `FAQPage`. - Prices, grades, and standard coverage are kept consistent across pages — cite the pack or benchmark page directly rather than paraphrasing figures from the homepage.