Aglet

Turn Regression Gate Results Into Better Agent Evaluation

Regression gates teach which evidence makes a release signal trustworthy. Preserve the crossing, protected slice, baseline, variance, and final action. Turn the case into a maintained gate example that separates agent movement from the evaluation system for the next release.

Keep the lesson for the next incident

  1. Save the gate case

    Store release change, baseline, slices, dimensions, scores, variance, evaluator, scoring, examples, reruns, outcome, and action. Include a borderline crossing and a measurement change so the review path is concrete for release reviewers.

  2. Improve gate practice

    Add pinned versions, protected slices, variance rules, raw outputs, borderline handling, and infrastructure classifications to evaluation plans. Explain how the change addresses the false block or missed regression observed in the protected slice.

  3. Watch for gate debt

    Set a signal such as repeated overrides, aggregate pass beside slice fail, missing baselines, rising borderline outcomes, or crossings after evaluator changes. Assign an owner to sample gate records and revise the contract when it appears.

What to carry forward

The learning record should connect the gate outcome to revised baseline, slice, or review practice and recurrence signal. State which evidence changed the action. Keep the lesson tied to the tested release gate rather than treating every crossing as a regression by itself.

Technical background: OpenAI evaluation guidance.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow