Approved Articulation Authoring
Continue deterministic accepted-only authoring and verify the exact saved USDZ joint graph. The workflow code, not the reasoning agent, performs mutation.
When to Use
Use only after the agent decision ledger and any configured human review receipt validate against the same candidate evidence.
Limitations
- Author exactly once and only accepted native-ready candidates.
- Do not author masses or colliders in the articulation stage.
- Render evidence cannot replace saved-stage identity and graph readback.
Prerequisites
Verify the decision ledger, reviewed candidate document, review receipt when required, and approved candidate document before continuing.
Instructions
- Let the deterministic controller build the accepted-only authoring request.
- Verify authoring result IDs match approved IDs exactly.
- Reopen the saved artifact and validate source identity and complete joint graph readback before rendering.
- Render focused validation evidence for the authored result.
- If validation fails, retain accepted work and route only failed candidate scope through targeted reinspection in a new run or supported refinement.
- Finalize
completedonly for exact readback; otherwise retain a conditional result with unresolved IDs and evidence.
Command Reference
After an agent patch with requested review policy none, standalone runs use
_agent-apply to continue deterministic authoring. An Asset
articulation.author.v1 leaf instead re-runs its same opaque
content-workflow-cli articulation agentic-leaf author --invocation ...
entrypoint with the exact immutable invocation. Do not import an internal
Python API or create or execute a Python helper to cross either decision pause.
Human-gated runs continue with the public articulation review command.
Common Workflows
In a composed asset workflow, the outer coordinator remains the only reasoning loop and re-invokes the selected Asset leaf's opaque entrypoint; it does not substitute the standalone prepare/apply surfaces.
Output Format
Retain approved candidates, receipt, authoring request/result, saved USDZ, validation evidence, checkpoint, progress, trace, and final summary.
Troubleshooting
On graph mismatch or artifact drift, keep the run for diagnosis and do not reauthor over immutable evidence. Start a new run after correcting inputs.