Question 220
A few-shot prompt uses examples from a different document type than production inputs. What should be tried?
- Remove all acceptance criteria and trust the examples.
- Add more detail to the instructions while keeping all examples from the wrong document type.
- Add more examples from the unrelated document type.
- Use representative examples that illustrate the actual task’s boundaries and expected outputs.
Explanation
Correct Answer
D:Misaligned demonstrations can teach the wrong patterns; examples should reflect the conditions under which the workflow is intended to perform.
Incorrect Answers
A:Examples complement explicit requirements but do not replace the need to define what the workflow must accomplish.
B:More detail may not correct conflicting demonstrations; the examples themselves should reflect the production inputs and expected decision boundaries.
C:Greater volume of mismatched demonstrations can reinforce the wrong assumptions instead of improving performance on real inputs.
Question 221
An employee asks Claude to paraphrase an entire paid report for redistribution. What should be considered?
- Paraphrasing guarantees the content is free of rights restrictions.
- Calling the result “AI-generated” grants publication rights automatically.
- Removing the author name resolves all concerns.
- Rights and permitted use; paraphrasing does not automatically authorize redistribution.
Explanation
Correct Answer
D:Changing expression does not by itself establish permission to reproduce protected content or distribute a substitute for the original report.
Incorrect Answers
A:A rewrite is not a license and cannot settle whether the extent and purpose of reuse are permitted.
B:A generation label does not override the rights or conditions attached to the source material being transformed.
C:Deleting attribution does not create rights to redistribute the work and can further misrepresent its provenance.
Question 222
A workflow checks whether a form contains three required fields before Claude drafts a response. What is sensible?
- Use Claude to guess whether absent fields probably exist.
- Remove field validation because the draft is fluent.
- Validate rules deterministically; use Claude for drafting.
- Ask the reviewer to discover all missing fields after sending.
Explanation
Correct Answer
C:Simple explicit rules can be enforced predictably, while the model addresses the part requiring flexible interpretation or drafting.
Incorrect Answers
A:Inference cannot establish whether required information was actually supplied and weakens a straightforward input control.
B:Readable output does not prove that the prerequisites required by the business process have been satisfied.
D:Late detection creates rework and may expose recipients to responses based on incomplete or invalid submissions.
Question 223
A brief says, “Escalate requests from new customers or orders over 5,000 with missing identification.” It is unclear whether missing identification applies to both groups. What should the prompt author establish before using this rule?
- Confirm the intended grouping with explicit conditions or boundary cases.
- Ask Claude to choose the interpretation that escalates the fewest requests.
- Add more examples without checking their approved outcomes.
- Require a confident classification without changing the rule.
Explanation
Correct Answer
A:The placement of the final condition changes which requests qualify. Parenthesized logic or agreed examples can resolve that material ambiguity.
Incorrect Answers
B:Minimizing workload does not establish the policy's intended scope and could exclude requests that require escalation.
C:Unverified examples can reinforce an arbitrary interpretation rather than establish what the rule is supposed to mean.
D:Confidence in the output does not remove the ambiguity in the condition Claude is being asked to apply.
Question 224
A Project’s collaborators change after a team reorganization. What maintenance is needed?
- Ask remaining users not to mention sensitive files.
- Change only the Project’s display name.
- Keep all historical members indefinitely.
- Review membership and permissions against current responsibilities.
Explanation
Correct Answer
D:Access should reflect present work needs, so former participants should not retain rights merely because they once contributed to the Project.
Incorrect Answers
A:Informal discretion does not change the actual access rights of people who no longer need the material.
B:Renaming the workspace does not revise collaborator permissions or align access with the new team structure.
C:Past participation does not establish a continuing need to view current knowledge or edit the maintained configuration.
Question 225
Stakeholders say a Claude-generated prototype looks clear, but no one has tried completing its main task. Which next step best tests usability?
- Ask stakeholders to rate the prototype's colors again.
- Treat positive comments as proof of task completion.
- Observe representative users completing the task without coaching.
- Have the designer explain every step while users click along.
Explanation
Correct Answer
C:Task performance reveals navigation and comprehension problems that favorable impressions alone cannot establish or rule out.
Incorrect Answers
A:Visual preference supplies little evidence about whether users can complete the intended workflow successfully.
B:Expressed approval is not an observation of successful use and may miss difficulties that appear only during interaction.
D:Coaching can conceal problems in the interface itself, so this does not establish independent task completion.
Question 226
Claude declares one survey result decisively better despite broad overlapping uncertainty intervals. What is needed?
- Remove the sample sizes from the report.
- Choose the result with the prettier chart.
- Ignore intervals because the averages are different.
- Account for uncertainty before claiming superiority.
Explanation
Correct Answer
D:Point estimates alone can overstate distinctions when sampling variability is large relative to the observed gap.
Incorrect Answers
A:Hiding sample information makes uncertainty harder to assess and does not justify a stronger conclusion.
B:Visual appeal does not evaluate statistical uncertainty or support the strength of the comparison.
C:A numerical difference between estimates does not itself establish a reliable difference in the underlying populations.
Question 227
A departing consultant supplies a complete workflow runbook. What should the receiving team do to verify that it can operate the Claude-assisted process independently?
- Confirm only that the runbook file opens.
- Ask the consultant to perform every step while the team watches.
- Approve handover based solely on the number of documented pages.
- Run a representative case and exception using the runbook.
Explanation
Correct Answer
D:A practical rehearsal tests whether the receiving team can apply the instructions, locate permissions, and resolve an exception without the consultant's intervention.
Incorrect Answers
A:File availability is necessary, but it does not establish that the documented operating instructions are sufficient or usable.
B:A demonstration can help training, but it does not verify that the receiving team can carry out the procedure independently.
C:Document length does not show whether the instructions contain the operational details needed to complete the task and handle exceptions.
Question 228
Claude reports “95% confidence” in an unsupported factual claim. How should that figure be treated?
- Treat the confidence number as stronger evidence than a source contradicting the claim.
- As proof that only one in twenty such claims is wrong.
- As permission to omit all supporting references.
- Unvalidated generated confidence, not calibrated assurance.
Explanation
Correct Answer
D:Unless validated through an appropriate method, a numerical confidence statement does not establish the probability that this specific claim is true.
Incorrect Answers
A:Generated confidence is not an evidentiary override and must not displace applicable authoritative information.
B:That interpretation assumes calibration which has not been demonstrated for the statement or workload.
C:A confidence label supplies no traceable evidence and does not remove the need for claim verification.
Question 229
A policy requires human review, but reviewers merely click approve without reading Claude’s recommendation. What is wrong?
- The nominal checkpoint is not providing meaningful oversight.
- The model should generate an approval on the reviewer’s behalf.
- The control is effective because a person clicked.
- Oversight can be removed because reviewers are busy.
Explanation
Correct Answer
A:A formal button does not ensure scrutiny; reviewers need time, information, competence, and authority to challenge or reject the output.
Incorrect Answers
B:Generated acceptance would misrepresent human review and bypass the accountability the policy is intended to preserve.
C:A human action alone does not show that the recommendation was evaluated against the relevant evidence and criteria.
D:Workload pressure does not establish that the control is unnecessary; capacity and workflow design should be addressed.
Question 230
Claude lists an exception in a footnote while the main instruction makes an unconditional promise. What should change?
- Keep the main promise because a footnote exists.
- Make the footnote text smaller.
- Delete the condition to improve flow.
- Make the material condition prominent and clear.
Explanation
Correct Answer
D:Important eligibility boundaries must be clear at the point of decision, not obscured by a misleading main statement.
Incorrect Answers
A:A technically present qualification may not cure an assertion whose ordinary reading materially overstates the policy.
B:Reduced visibility increases the risk that readers miss the boundary governing the headline claim.
C:Removing a real exception makes the promise even less faithful to the source and less useful operationally.