A benchmark for agents that run entire medical research workflows.
The benchmark's stage rubrics become a teacher that post-trains the agent.
Verifiable environments where the model and its harness co-evolve.
