Evals

Quality you can
point to.

An Eval is a stored quality gate: criteria, thresholds, and an LLM-as-judge that scores the artifact and returns a verdict with a full report. Low-quality work stops before it ships.

Apply — Free Review

What an Eval is.

A declared EvalSuite with criteria, pass thresholds, and a judge that produces a verdict and evidence report.

  • Scores artifacts against declared criteria
  • Uses an LLM-as-judge, not a checklist
  • Pass modes: all criteria or a minimum score
  • Returns a verdict and a full reason report
  • Passed eval report is required evidence
  • Registered in the evals registry

Exactly how it runs here.

Evaluations apply declared criteria and return evidence-backed reports. The Content Department uses a copy quality gate before campaign work is marked ready.

How an Eval works.

  1. 1

    A Department registers an EvalSuite with criteria and thresholds

  2. 2

    The run produces an artifact

  3. 3

    The judge scores it against the criteria

  4. 4

    The verdict and report are stored with the run

  5. 5

    The artifact advances only on a passing report

Want this discipline in your company?

The Agent Brief maps one of your workflows to the same loop, approvals, and evidence.

Apply — Free Review