See where simulation matches execution and where important failures are missed. This research preview helps teams inspect prediction quality with claim-scoped evidence.
Compare model predictions with executed robot trajectories under a frozen protocol and review where they diverge.
Inspect which failure patterns a model notices and which it misses, so assurance claims stay scoped to the evidence.
Use live MotionBench for paired session comparison when both runs exist. Use this research preview when you need claim-scoped prediction scorecards.
Built from V7 trained-proxy research. Scorecards are claim-scoped. Coverage methods that fail validation are excluded from release authority. Details: /benchmarks#v7-world-model.
Explore this research preview and the V7 card on /benchmarks. Authenticated users can open the dashboard world-model viewer.
Treat it as research evidence for assurance conversations. Production promote or block decisions should use validated evaluation workflows.
Loading Arisole. If the interactive app does not start, continue with the public links above.