Aglet

Learn From Repeated Missing Evidence in Agent Work

Learning from missing evidence means improving how work asks for, records, and cites observations. Preserve the claim that failed review, the smallest artifact needed, and the uncertainty that should have been stated. Use those lessons to make future evidence requirements visible before execution.

Keep the lesson for the next incident

  1. Classify evidence failures

    Group incidents by absent source, stale artifact, omitted comparison, unclear provenance, unsupported inference, incomplete cohort, or handoff omission. Keep one concrete claim and review consequence for each class. Distinguish a collection failure from a claim that needs narrower scope.

  2. Keep evidence exemplars

    Retain short claim ledgers and handoff examples showing source, observation time, transformation, comparison, uncertainty, and reviewer action. Include a deliberately unsupported claim with the correct limitation language. Update examples when source types or decision boundaries change.

  3. Review recurrence and ownership

    Assign owners for evidence collection, claim review, and handoff quality when they differ. Sample material results on a defined cadence and track repeated gap classes. Reopen investigation for an unsupported claim, stale source, or missing comparison that appears after the evidence checklist was applied.

What to carry forward

Close the learning record with evidence-gap classes, claim-ledger examples, owners, and a recurrence trigger tied to unsupported material claims or stale sources. Preserve accepted limitations as part of the workflow. The durable result is more reviewable reasoning, not a promise that every task has complete evidence.

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