Lab 02
Governed AI Run Simulator
Explore how an evidence-grounded workflow moves from a fictional brief to a controlled publishing decision.
Interactive reconstruction · No live model calls
This is a deterministic educational simulation with fictional fixtures. It is not a live production system: no model generates anything, the reviewers and verifier are simulated, and nothing is published.
Operate a run
Status Choose a fictional scenario, then start the run.
Publish-gate decision
NOT DECIDED
The gate decides only after every step has run or been skipped.
Inspect the run
Fictional brief
What the run was asked to produce
Pre-written fictional draft
Claim-to-source mapping
Every material claim and the sources behind it
| Claim | Supported by | Contradicted by | Status |
|---|
Fictional sources
Findings
Quality, policy and independent-review findings
Authoritative run history
The engine's ordered, append-only record
The page only displays this history. Each event is chained to the one before it, so an edited or reordered history is rejected.
Final gate explanation
Why the gate decided what it did
How the gate decides
- BLOCKED if any required condition failed or did not complete, including a verification timeout.
- NEEDS HUMAN REVIEW if nothing failed but a contradiction or a reviewer disagreement is open. The gate never resolves it on anyone's behalf.
- READY only when all eight required conditions passed. Publishing remains a human decision.
Unknown scenarios, invalid fixtures and edited run state also end as BLOCKED.