One evidence spine, two actors · the physical actor

Watch a verified plan execute.

Autonomous warehouse robots fulfil orders, drive around the pods, and yield to each other and a human — collision-free and people-first by construction. Origin’s deterministic oracle (finish / escalate / refuse) verifies every plan, a deterministic coordinator keeps every executed step safe, and the whole run drops a signed Origin Attestation you re-verify offline — the same evidence spine a digital agent action earns.

Everything runs client-side and is seeded + replayable. The environment is the moat, not the model: robots follow the oracle-verified plan — no learned-policy result is claimed. A verdict means "reproducible under this verifier," never "safe." Synthetic layout.