Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Free-form CAD reconstruction
Each model built a converter that turns calibrated views and sections of pump and flow components into STEP solids, nominal and twisted or bent; each output is drawn against the true solid. Hidden test parts are the ones the reward is computed on (three components in three variants, for each model's best and median attempt); public dev parts cover every attempt, but agents could test on them, so they are not graded.