How to use it
- Open Expectations.
- Click New Expectation.
- Choose an Agent.
- Write the expected behavior in Expectation.
- Click Save and Run.
- Use Run, Edit, Disable, Enable, or Delete to manage existing expectations.
- Use Filter expectations… and the sort menu to find expectations by result or recency.
How to interpret the results
The page summarizes Total expectations, Expectations with violations, and Sessions evaluated. Each expectation shows one of the statuses used in the product:- Passing: no violations are currently shown.
- Violating: at least one evaluated trace failed the expectation.
- Disabled: the expectation is turned off and will not run.
Next steps
- Inspect evidence in Behavior Search.
- Rerun a failing turn with Replay.
- Use Clustering to find recurring patterns that may need expectations.