Testing Center subagent and action evaluation
Key Takeaway
Subagent assertion and action assertion are default evaluations that score each test case 1 (pass) or 0 (fail), unlike optional response quality evaluations, which an LLM judge scores on a 0–5 scale. If the agent picks the right subagent but misses an expected action, the test case scores 1 on subagent assertion and 0 on action assertion.
Agentforce Testing Center runs batches of test utterances against an agent and scores each result. Two of its default evaluations check the agent's choices rather than its wording. Subagent assertion checks whether the agent picked the right subagent. Action assertion checks whether the agent picked the right actions. Their pass/fail scores show where the agent's routing or action selection breaks, so you can fix the agent before launch.
Default evaluations and response quality evaluations
Every Testing Center test runs three default evaluations: response evaluation, subagent assertion, and action assertion. Default evaluations always run.
You can also add optional response quality evaluations, such as completeness, coherence, conciseness, latency, and instruction adherence. The two kinds of evaluation are scored differently.
| Default evaluations | Response quality evaluations | |
|---|---|---|
| When they run | Always | When you add them |
| Score | 1 (pass) or 0 (fail) | 0–5, from an LLM judge |
| Passing score | 1 | 3 or higher |