Testing, Deployment, and Maintenance10% of the exam
Testing Center evaluations and test results
What each Testing Center evaluation compares, how default, quality and custom scorers are judged and scored, why non-deterministic agents still need evaluation at scale, and how to read a run and act on failed test cases.
Explain how Testing Center evaluations work.
~7 min read · 8 practice questions ·
What the exam expects
- Explain what each default evaluation compares: subagent, actions and response, expected against actual.
- Predict the result when an agent runs only some of the expected actions.
- Name the quality scorers (completeness, coherence, conciseness, latency, instruction adherence) and say what custom scorers add.
- Explain why non-deterministic output is a reason to evaluate at scale, not a reason to skip it.
- Read a finished run and choose the right fix for a failed test case: change the agent or change the test.
- Key terms: evaluation / scorer (a check run on every test case; the Studio Testing Center says scorer); assertion (a pass or fail comparison of expected and actual subagent or actions); LLM-as-judge (one model scores another model's output against criteria); expected response (the answer you expect the agent to give); pass rate (the share of test cases that passed one evaluation in a run).
Key facts
Default scorers
- Detail
- Turn-level: response, subagent, action; conversation-level: task resolution, subagent, actions; always run
Default evaluation scores
- Detail
- Subagent and action: 1 pass or 0 fail; Response Evaluation: LLM-judged, 3 to 5 generally succeeds
Action assertion
- Detail
- Fails if any expected action is missing or a wrong action is chosen
Conversation-level actions
- Detail
- Pass if all expected actions occurred, extra actions allowed
Response evaluation
- Detail
- LLM judge compares agent response with expected response
Expected response content
- Detail
- The response you expect, not instructions about it
Quality scorer threshold
- Detail
- 0-5 scale, 3 or higher passes
Custom scorer output
- Detail
- Labels (with fallback label) or free-form; you set the threshold
Run overview
- Detail
- Subagent, Action and Response Evaluation Pass %
Sources for this section (5)
- Default Scorershelp.salesforce.com (opens in a new tab)
- Use Test Resultshelp.salesforce.com (opens in a new tab)
- Choose Evaluationshelp.salesforce.com (opens in a new tab)
- Create Custom Scorershelp.salesforce.com (opens in a new tab)
- Understanding Agentforce Testing Center Resultstrailhead.salesforce.com (opens in a new tab)
- What does each default evaluation compare?
- Why can a correct answer fail response evaluation?
- When should you add a custom scorer?
- Why does non-determinism make evaluation at scale more important?
- How are quality scorers such as completeness scored?
The questions are free to read. Answers with their sources and self-grading are included with the pack.
Keep reading with the Agentforce Specialist pack
- Core concepts
- Exam traps
- Worked scenario
The pack includes:
- All 26 lessons of the Agentforce Specialist guide, with reading progress
- Interactive decision walks and step-by-step sequence steppers
- Inline practice questions and one-click topic practice (8 questions for this lesson)
- An official source cited for every claim
- An AI tutor that answers from the guide
- Unlimited timed practice exams