An agent can do work. This decides whether it counts.
A real governed repository, a real task with real path authority, real verification and evidence, and your approval as the human owner — followed by the two refusals that make the point: correct work outside its authority is blocked, and a previously valid approval goes stale the moment the governed world changes. Everything you see is the product's own governor evaluating this run's own repository.
- Project
- →Run
- →Work
- →Boundary
- →Evidence
- →Review
- →Approval
- →Advance
- →Violation
- →Recovery
- →Re-approve
- →Stale
- →Done
An agent can do work. The governor decides whether it counts.
Prairie Service Portal is a deliberately tiny customer-intake app. The repository you are about to control is real: the real ge init created it, and the real Governed Engineering governor walked it — with reviews, approvals, and a declared work graph — to the implementation phase. A coding agent (deterministic in this demo — no AI writes code here) will do one small task. Governed Engineering decides, from evidence, whether the project may advance.
You will act as demo-owner, the human whose approval the advancement ceremony consumes. Demo identities are fabricated and run-scoped; nothing here claims authenticated real-world intent.