AI Evaluation Lab
Independent, provider-neutral evaluation and reproducible evidence for AI systems and agents.
The Lab develops task contracts, independent verifiers, model and agent evaluations, and evidence that other people can inspect and reproduce.
Explore the main repositoryResearch roadmap
Research programs
Applications built on the Lab
ProofGate
A coding-agent verification application built on RARB for the Nebius × NVIDIA hackathon. Run Python verifier checks against recorded candidates, inspect failed gates, and explore the preserved evidence.