Pick an environment on the left, then a run — step through what the agent saw, did, and was scored on completion, efficiency, accuracy, and safety. Reward-robustness runs also show a proxy-vs-oracle verdict; hacks are flagged.