Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Bearing fault embeddings
Each model's embedding of 2,048-sample vibration records, seen through the task's fixed linear probe: hidden test records come from eight unseen machines and are the ones the reward is computed on (each model's best and median attempt). Public dev records come from the recorded fleet, but agents could test on them, so they are not graded.