Module 07 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. Why run both a step ceiling and a repeat detector?
- A. for redundancy
- B. the repeat counter catches tight loops early; the ceiling catches drift that never repeats exactly
- C. one is for cost, one for time
- D. frameworks require both
2. An agent that emails a customer when it was asked a question is…
- A. misbehaving and needs a stricter instruction
- B. acting within the capabilities it was given - a permissions problem
- C. correctly being helpful
- D. hallucinating
3. A high tool-invention rate means…
- A. the model is weak
- B. the goal implies capabilities the tool list does not contain
- C. the prompt is too short
- D. the registry is misconfigured
4. Staleness budgets should be…
- A. one global number
- B. per fact, based on how often that kind of fact changes
- C. set to one day
- D. handled by re-running the whole workflow
5. A confident wrong answer with no reported error and no empty output is…
- A. caught by the error rate
- B. invisible to operational monitoring, and detectable only with a case set of known answers
- C. rare enough to ignore
- D. caught by the loop detector
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - the repeat counter catches tight loops early; the ceiling catches drift that never repeats exactly
A ceiling of eight lets a two-step loop run four times - eight paid calls to learn something detectable at the third.
2. B - acting within the capabilities it was given - a permissions problem
An agent whose job is to answer should hold no tool that sends email.
3. B - the goal implies capabilities the tool list does not contain
Narrow the goal or add the tool deliberately. Another sentence listing the tools does not address it.
4. B - per fact, based on how often that kind of fact changes
A single rule either re-fetches contract terms constantly or uses month-old balances.
5. B - invisible to operational monitoring, and detectable only with a case set of known answers
It is the most expensive failure and the only one that announces nothing.
Practical Check
Apply this module to your own work: complete the module activity for *Failure Modes: Loops, Over-action, Hallucinated Tools and Stale Context*, 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.