The trust score is Critiqor’s top-level reliability signal for a run. It reflects the quality of your agent’s observed runtime behavior — not just its final output. After each critiqor finalize, the run report includes a trust level alongside the primary diagnosis and recommendations. This page explains what each trust level means and how to put trust scores to work.
Trust levels
High
The run looks healthy. No significant reliability issues were observed during the session. Your agent completed the task without notable failures, excessive retries, or cost inefficiencies.
Moderate
The run completed, but showed signs worth reviewing. Something in the runtime evidence deserves attention — check the primary diagnosis and recommendations in the dashboard to understand what was flagged and whether it warrants a fix.
Low
Significant reliability issues were observed during the run. Review the primary diagnosis and the supporting evidence before relying on the results of this run. A Low trust score is a strong signal that your agent’s behavior needs investigation.
How to use trust scores
- Quick health check — after each run, glance at the trust level to get an immediate read on whether the session went well.
- Track regressions over time — compare trust levels across runs as you make changes to your agent. A trust level that drops between runs is a clear signal something regressed.
- Start diagnosis with Moderate and Low runs — use the dashboard to open the run, read the primary diagnosis, and review the evidence that drove the trust level down.
- Don’t rely on trust scores alone — trust levels are a starting point, not a complete picture. Always read the primary diagnosis and recommendations to understand the context behind the score.
Trust score in deployment checks
If you want to gate deployments or CI steps on reliability, critiqor check lets you assert a minimum trust level before proceeding. See Policy Checks for configuration details.
Trust levels are reliability signals, not guarantees. A High trust score does not mean your agent is production-ready — use it alongside your own testing and review process.
