Evaluations
—Loading durable stateActive rollouts
—Queued + leased + gradingCompleted rollouts
—Evidence persistedRecovery state
ReadyLease-based resumeCreate evaluation
Run a reproducible policy suite
Durable queue
Evaluations
| Evaluation | Policy / suite | Backend | Progress | Status | |
|---|---|---|---|---|---|
| Connect to load evaluation state. | |||||
Runtime inputs
Registered policies
No policies loaded.
Pinned tasks
Task suites
No suites loaded.
Evidence routing
Three judges. One deterministic truth.
Simulator predicates determine success. Purpose-built video analysis and independent review explain exactly what happened.
Simulator metrics→Robotics ER 2→Gemini 3.7 · High→GPT‑5.6 Sol · Max