01
edith-1
Introducing edith-1, a decision model for detecting reward hacking and other failure modes in agent trajectories and environments across production, evaluation, and RL rollouts.