Current Testing Surface
- Use
/testing/ab-testfor live traffic split testing between a base agent and a test agent. - Treat
/testing/evaluationas a legacy route. It exists in code, but it is currently unused in product flows.
Run An A/B Test
- Open
/testing/ab-test. - Create a test in the
Plannedcolumn. - Set the category, end criteria, traffic split, and success criteria.
- Edit the generated test agent.
- Start the test.
- Monitor
Planned,In Progress, andCompletedstates.
Legacy Evaluation Route
- Only reference
/testing/evaluationif the user explicitly asks about legacy evaluation workflows. - Do not route general testing questions to
/testing/evaluation.
Rules
- ALWAYS keep the A/B variant limited to the intended change.
- NEVER present
/testing/evaluationas the primary testing path in the current product. - IF the task requires live call traffic split, THEN use A/B testing.

