Mission Review Checklist

Classify the run before deciding what it proves: offline file rehearsal, browser fixture lab, provider commissioning, full-stack rehearsal or production execution.

1. Offline theatre evidence

• Read report.json: schema_version: 2, execution_mode: offline-fixture, six documents and no input issues.
• Confirm live_provider, browser_executed, settlement_approved and production_approved are all false.
• Verify all five evidence artifacts and recompute the acceptance checks:

  python3 demo/astral-omnidominion-operating-system-command-theatre/run_demo.py --verify-report reports/astral-omnidominion-operating-system-command-theatre/report.json

• Inspect accepted separately from integrity. An intact rejected bundle is valid evidence of rejection.
• Check the source, duplicate, deliverable and separate checker: four rows → three unique entries → 21,999 cents.
• Demonstrate --scenario rejected (one-cent error, exit 1) and --scenario paused (no task artifacts, exit 1).
• Open the HTML dashboard, inspect the job catalog and confirm no success card implies live capability or settlement.
• Retain the report and its unique runs/ directory. Hashes are relative to this report, not a signature or external attestation.

2. Full-stack artifact integrity

• Read reports/agi-os/first-class/first-class-run.json → steps[]. Require relevant stages to have status: success; enumerate skipped and failed stages.
• Recompute a summary hash with a cross-platform command:

  python3 -c "from pathlib import Path; import hashlib; print(hashlib.sha256(Path('reports/agi-os/grand-summary.md').read_bytes()).hexdigest())"

  Compare it to first-class/first-class-manifest.json. Verify the complete manifest before relying on the package.
• Confirm outputs belong to the run's commit and configuration; stale files cannot fill in a skipped stage.
• Inspect grand-summary.html, owner-control-matrix.json and the preserved first-class/owner-control-map.mmd using a trusted Mermaid viewer. A source diagram is not deployed-state proof.
• Review mission-bundle/manifest.json, simulation telemetry and dry-run.log when present. Completion markers do not turn a simulation into live evidence.

3. Owner authority and live workers

• Follow the owner field guide; verify actual network, contract addresses, owner and pausers.
• Preview only supported governance actions; use the actual pause script rather than report-generator subcommands.
• If a transaction was authorized, retain its receipt and re-query each relevant module. Do not assume instantaneous finality or UI refresh.
• For live work, collect the commissioning evidence, including isolation, enforced limits, stop/recovery, independent review and settlement prerequisites.
• Reconcile unknown provider outcomes from durable journals. Do not retry by deleting a dispatch record.

4. CI, archive and decision

• Run the targeted checks and inspect the required checks for the exact PR commit.
• Archive only reviewed, non-secret evidence. For example, package the offline output separately from the full stack:

  tar -czf astral-offline-evidence.tar.gz reports/astral-omnidominion-operating-system-command-theatre

• Record what passed, what failed, what was skipped and what remains uncommissioned.
• Use precise status: “offline rehearsal passed” or “provider commissioning passed for these tasks.” Reserve production approval for the actual deployment and its evidence.

The $40T/year figure and catalog budgets remain planning assumptions. Document-length scores and generated diagrams are not production acceptance gates.