Aglet

Investigate Why Agent Recommendations Conflict

Investigation should explain how two recommendations reached different actions for a shared question. Replay matched contexts where possible, trace evidence and intermediate reasoning, and vary one input or criterion at a time. Keep genuine policy disagreement separate from missing context or inconsistent execution.

Build a useful investigation brief

  1. Pair comparable runs

    Select runs that answer the same normalized question and record task text, source set, time window, revision, constraints, and output. Exclude or label runs with different objectives. Compare the first claim or intermediate value that diverges, not only the final recommendations.

  2. Trace decision inputs

    Follow each recommendation to source observations, evidence selection, calculations, thresholds, assumptions, and criteria. Check missing artifacts, stale context, different scope, and wording changes at each boundary. Record exact input and output revisions so another reviewer can reproduce the split.

  3. Vary one condition

    Hold the shared context constant while changing one evidence item, criterion, time window, or instruction that might explain the conflict. Compare actions and confidence language. If both options remain valid under the stated objective, document the policy decision needed rather than forcing convergence.

What to carry forward

Produce a conflict investigation with matched runs, first divergence, evidence lineage, tested conditions, and decision effect. Recommend a shared-input correction, criterion clarification, or human policy choice. Do not call one recommendation correct without proving its question, evidence, and objective matched the alternative.

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