---

name: parity-and-tooling description: "Guides RoboVerse cross-simulator parity experiments, registration and environment diagnostics, rollout tooling, and safe source-script selection." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Parity and Tooling

Use this route when comparing observations, rewards, policies, or rendered rollouts across simulators; auditing registration; choosing a conversion or rendering diagnostic; or deciding whether a repository script can safely become a reusable helper.

Route

1. Read parity-workflows.md and define the comparison target, aligned initial state/action/seed, selected backends, metric, and stopping rule before running anything.
2. Run the safe registration/environment check in scripts/verify_registration.py for import and discovery diagnostics. It does not prove simulator parity.
3. Use script-selection.md to classify scripts as adapt, reference-only, or exclude. Do not run real-robot, credentialed, long-training, sweep, or destructive conversion scripts as smoke tests.
4. Report measured deltas and backend identity. Distinguish observation/reward agreement from closed-loop policy transfer and from visual similarity.
5. Use troubleshooting.md for missing backends, mismatched state/action order, renderer failures, and misleading parity.

Parity measurements should be reproducible and honest: a clean number must not hide that a task fell into the void or that both sides used different resets.