Check whether the outcome improved
Set coordination checks
Require role, input, output, owner, handoff, message, dependency, shared-state, conflict, recovery, and completion rules. Define acceptable serial and parallel paths. Mark unobserved transitions as unverified for scenario review with all roles visible.
Review team variants
Apply checks to serial, parallel, conflicting, delayed, missing-message, failed-agent, and recovery cases. Inspect messages beside state updates and outputs. Record whether synthesis hides duplicate or missing work from independent reviewers.
Approve the scope
State which roles, team sizes, task types, and paths the result supports. Record fixture, versions, owner, reviewer, and retest trigger. Mark results inconclusive when shared state or message order cannot be observed.
What to carry forward
The verification record should show roles, messages, dependencies, state changes, conflict handling, recovery, and limitations. Approve only the coordination behavior observed. Reopen it when roles, communication, shared state, or task mix changes. Record a follow-up when ownership or message evidence remains disputed before accepting a combined result.
Technical background: Agent evaluation research on arXiv.
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