Assurance
Technical evaluations
A Technical Evaluation measures AI behaviour against a defined method, dataset and threshold. You run it in your own environment and submit signed results to AIGovernr.
- In the app
- Project → Understand & Assure → Evaluations
Define the gate
- 1Choose Create gate requirement and pick an evaluator from the Evaluator catalog, such as performance, fairness, robustness, privacy or prompt injection.
- 2Set the dataset, metric and threshold, and choose Approve new specification version.
- 3Choose Queue for external runner, or use CI config to generate a GitHub Actions workflow for your pipeline.
AIGovernr does not execute evaluations itself. Your runner posts an HMAC-signed result, which keeps the result tamper-evident.
Interpret results
- Each submitted result is stored as evidence and shown in Run history. Choose Check gate to see whether the latest result passes.
- A failed gate raises a High-severity Finding, linked to a Risk when one is set on the requirement.
- Use Review nuanced finding and Record attestation when a result needs human judgement.
Last reviewed 2 October 2026
Next stepDPIAA Data Protection Impact Assessment reviews processing that may create a high risk to people's privacy. In AIGovernr each DPIA belongs to one AI System.