Module 06 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. Evaluating agents requires looking at the trajectory, not just the final answer, because…
- A. It is more data
- B. A correct answer reached via an unsafe or wasteful path is still a failure
- C. Answers are hard to grade
- D. Tools are slow
2. Which agent metric matters most for safety?
- A. Token count
- B. Whether consequential actions were correctly gated for approval
- C. Number of steps
- D. Latency
3. Measuring 'stopping behaviour' means checking that the agent…
- A. Runs fast
- B. Terminates at the right point rather than looping or halting prematurely
- C. Uses few tools
- D. Returns JSON
4. Unnecessary tool calls are worth tracking because they…
- A. Look untidy
- B. Raise cost and latency and widen the surface for injection and error
- C. Break schemas
- D. Confuse users
5. A good agent eval case includes…
- A. Only happy paths
- B. Cases where the correct behaviour is to refuse, escalate, or stop
- C. Only long tasks
- D. Only tool tasks
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - A correct answer reached via an unsafe or wasteful path is still a failure
For agents, how the result was reached is part of correctness.
2. B - Whether consequential actions were correctly gated for approval
A skipped gate is a direct path to an irreversible real-world effect.
3. B - Terminates at the right point rather than looping or halting prematurely
Both non-termination and premature exit are common and separately damaging.
4. B - Raise cost and latency and widen the surface for injection and error
Every avoidable call is avoidable risk as well as avoidable spend.
5. B - Cases where the correct behaviour is to refuse, escalate, or stop
If you never test refusal, you have not tested the control you rely on.
Practical Check
Apply this module to your own work: complete the module activity for *Agent Evaluation: Tool Trajectory, Approvals, and Stopping*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.