AIGRCeler

Examples

Everything on this page is fictional. The company, the people and the events are invented. The records are real output of AIGRCeler.

The dashboard page for one AI system in the example run, TalentLens: where it is in the four stages, and a go decision that reads Not ready, with the controls and the open incident that stand in the way.
One AI system in the example run, in the dashboard AIGRCeler builds. It shows where the system is in the four stages and what stands between it and a go decision. The run was stopped early on purpose, so the answer is "Not ready". The company and the people are fictional.

Brightwell Talent Partners

An invented staffing company. Its AI hiring screener, TalentLens, is a feature of its applicant tracking system that the vendor switched on without anyone at the company asking for it.

On 14 October 2026 the run set up the company, went through Discover, Assess and Govern, and started Operate. Along the way it recorded a stage block, three temporary measures, an incident, an obligation to an outside client and the approval of 75 controls. Invented people answered in everyday words, and AIGRCeler did the translating into its records.

Open the example on GitHub

What you can read

Decision journalOne line per decision: who decided, when, on what basis.
AI systemsThe register of systems and their owners.
Company risksRisks that belong to the company, not to one system.
DashboardThe governance dashboard built by the engine.
The product itselfThe folder you can open in Claude Code and continue.

What to keep in mind

  • Not a result for your company. It is one invented case. It shows how AIGRCeler builds and keeps records over time, not that it will give the same outcome for a real company. Nothing in it is legal advice.
  • A pre-release build. The run used a build made before the first release. It has the same four stages and the same 104-control starter catalog as the released product, but it differs in details. A new run on the released product is planned.
  • Only one system was run. Strong machine-made evidence, a tampered evidence file, a change of the vendor's model version, a risk acceptance, a closed incident and a first review were not reached. The example lists each one.
  • Not needed to install. The examples live in a separate repository and have no releases of their own.