Aglet

Investigate Missing Evidence in an Agent Recommendation

Investigation should show exactly where a recommendation outruns its evidence. Decompose the result into claims, trace each to source observations and transformations, and compare stated conclusions with the artifacts available at the time. Keep missing, stale, contradictory, and irrelevant evidence distinct.

Build a useful investigation brief

  1. Build a claim ledger

    List every material claim, its type, required evidence, cited artifact, observation time, and current status. Mark direct observation, derived calculation, assumption, interpretation, or unsupported assertion. Include rejected or omitted artifacts so the ledger reflects the full review boundary.

  2. Trace evidence lineage

    Follow each cited value from source artifact through filtering, aggregation, transformation, and final wording. Check identifiers, timestamps, revisions, and selection criteria. Record where evidence is lost, summarized without support, or replaced by an unstated assumption.

  3. Test the decision boundary

    Compare the recommendation with a small alternative case, counterexample, or missing observation that could change the outcome. Identify which claims move and which remain stable. If no available artifact can test the boundary, state the unresolved question and required source.

What to carry forward

Produce a claim-and-evidence brief with lineage, provenance, omitted artifacts, tested boundaries, and unresolved gaps. Recommend a bounded evidence request or a narrower conclusion. Do not upgrade an unsupported assertion because its surrounding context appears plausible.

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