Anthropic CCAO-F (Claude Certified Associate - Foundations) Practice Test 1

Question 1

A coordinator asks Claude to summarize a meeting. The result is readable but omits owners and deadlines. The transcript contains both. Which revision most directly addresses the failure?

  1. Specify a table with action, owner, and deadline columns, using “not stated” for missing values.
  2. Request a shorter summary with fewer details.
  3. Ask Claude to adopt the role of a senior coordinator without changing the requested output.
  4. Ask for a more professional writing style.

Explanation

Correct Answer

A:

An explicit output contract makes the required information visible and checkable against the transcript. The missing-value rule also prevents the model from filling gaps with plausible names or dates.

Incorrect Answers

B:

Compression may remove exactly the operational details that matter. Brevity is useful only after the required fields and handling of omissions are specified.

C:

Role framing can provide context, but it leaves the deliverable ambiguous. Defining the required fields is more direct than expecting a role description to imply them.

D:

Tone instructions influence presentation, but do not identify the fields the coordinator needs. A polished narrative can still omit responsibility and timing.

Question 2

Claude cites a genuine report for the claim that a pilot reduced processing time by 40%. The linked passage discusses the pilot but contains no measured time reduction. What should a reviewer conclude?

  1. The citation does not substantiate the numerical claim; locate supporting evidence or remove the number.
  2. Accept the number after finding another generated answer that repeats it.
  3. The claim is verified if the report comes from a reputable publisher.
  4. The claim is verified because the report exists.
  5. The claim can be retained after changing 40% to approximately 40%.

Explanation

Correct Answer

A:

Verification requires the evidence to support the specific assertion, including its magnitude and scope. A related passage is insufficient, so the number must remain unverified until directly supported.

Incorrect Answers

B:

Agreement between generated responses does not provide independent evidence that the cited report contains the claimed measured improvement.

C:

Publisher quality helps assess credibility, but claim-level support is still required. A credible report can be cited for a conclusion it never makes.

D:

Source existence establishes that the reference is real. It does not establish that the source contains evidence for the particular statement being attributed to it.

E:

Adding an approximation changes precision without supplying evidence. An unsupported estimate remains unsupported even when qualified by softer wording.

Question 3

A team needs the same writing rules applied in every new chat within a Claude Project. Its approved reference documents are already in project knowledge. Where should the recurring writing rules be placed?

  1. Only in the opening message of one existing chat.
  2. Only in the project title.
  3. In project instructions.
  4. Place the rule only in one generated output and expect future chats to infer it.
  5. In the filenames of the reference documents.

Explanation

Correct Answer

C:

Project instructions provide reusable guidance for chats in that project. They are appropriate for recurring rules such as audience, tone, formatting, and handling of missing information.

Incorrect Answers

A:

A message in one conversation is not a reliable way to establish guidance for every separate conversation. Reusable project behavior should be configured at the project level.

B:

The title identifies the workspace rather than defining a detailed writing policy. A meaningful name helps organization, but does not replace instructions.

D:

An output example may illustrate style but is not a maintained instruction applying explicitly across the Project’s later conversations.

E:

Descriptive filenames help identify documents, but are a poor substitute for explicit behavioral guidance. Encoding a style policy in names makes it ambiguous and hard to maintain.

Question 4

An approved Claude workspace will help summarize customer complaints. Names, phone numbers, and payment details are unnecessary for the analysis. Which input preparation best follows data minimization?

  1. Remove customer names but retain phone numbers as convenient identifiers.
  2. Place a confidentiality warning above the unmodified export.
  3. Remove unnecessary personal and payment fields before supplying the complaint text.
  4. Keep all fields because more context is always useful.

Explanation

Correct Answer

C:

The task requires complaint content, not customer contact or payment information. Removing unused sensitive fields reduces exposure while retaining the information needed for the approved analysis.

Incorrect Answers

A:

Phone numbers can still identify people. Removing only one identifier leaves unnecessary personal information in the material supplied for analysis.

B:

A warning communicates expectations but does not reduce the information disclosed. The safer preparation step is to omit fields the task does not need.

D:

Additional data can increase exposure without improving the task. Input selection should follow the analytical purpose rather than assume that every available field is beneficial.

Question 5

A support workflow uses Claude to draft refund recommendations. Company policy requires an authorized employee to approve every refund before money moves. Which design meets that requirement?

  1. Create a draft recommendation and require explicit authorized approval before the payment action.
  2. Replace approval with a daily count of completed refunds.
  3. Let Claude approve refunds whenever its explanation sounds confident.
  4. Execute the refund first and send the recommendation for later review.

Explanation

Correct Answer

A:

Separating drafting from execution creates a control point before the consequential action. The workflow should record the authorized decision and execute only the approved transaction.

Incorrect Answers

B:

Aggregate monitoring supports oversight but does not approve individual transactions. A daily report leaves the prohibited execution path open.

C:

Confident language is not authorization and does not establish correctness. The policy assigns approval to an employee, so model confidence cannot substitute for that decision.

D:

Review after execution cannot enforce a prerequisite that must occur before money moves. It may detect problems, but it does not satisfy the stated approval policy.

Question 6

An internal report says revenue rose from 80,000 to 100,000. Claude describes this as a 20% increase. How should the reviewer verify and correct the statement?

  1. Report a 25,000 increase because the percentage difference is 25.
  2. Divide 20,000 by the initial 80,000 and report a 25% increase.
  3. Average the two values before calculating the percentage.
  4. Accept the percentage because the absolute difference is correct.
  5. Divide the difference by the final value and retain 20%.

Explanation

Correct Answer

B:

The change is 20,000, and the baseline is 80,000. Recomputing the arithmetic independently yields 0.25, revealing that the generated percentage used the wrong reference amount.

Incorrect Answers

A:

The correct relative rate is 25%, but that percentage does not change the actual absolute difference of 20,000 between the supplied amounts.

C:

The midpoint is useful for some symmetric percentage-change measures, but that is not the ordinary increase from an initial value to a final value. The report gives no instruction to use a midpoint convention.

D:

A correct absolute difference does not validate the rate. Both the numerator and denominator must match the intended comparison before the percentage is accepted.

E:

Using the final value as the denominator answers a different comparison. Percentage increase is measured relative to the starting amount unless another basis is explicitly defined.

Question 7

A team can choose among several approved Claude models. It needs short document classifications within a latency target and a defined accuracy threshold. What is the strongest selection method?

  1. Choose by model name alone without testing the classification task.
  2. Use whichever model produced the most confident answer in one demonstration.
  3. Select the model with the longest generated explanations.
  4. Evaluate representative labeled examples and select a model that meets both quality and latency requirements.

Explanation

Correct Answer

D:

Task-specific evaluation measures the outcomes that determine suitability. Representative examples and explicit acceptance thresholds support a defensible choice without assuming that model reputation predicts performance on this workload.

Incorrect Answers

A:

Names and general positioning can narrow candidates, but cannot establish performance on the actual data. The task requires measured quality and response time.

B:

A single demonstration is too narrow to characterize reliability. Confidence in the wording also does not measure label accuracy across the intended workload.

C:

Explanation length is not the classification objective. It can add latency while providing no evidence that labels are accurate or that the response deadline is met.

Question 8

A revised Claude prompt produces better results on two examples but worse results on several others. The team changed the prompt, reference documents, and model simultaneously. What should it do next to identify the cause?

  1. Hold the evaluation set and other settings fixed while changing one factor at a time.
  2. Use a different set of examples for every trial.
  3. Compare only the best output from each configuration.
  4. Keep changing all three together until the outputs look consistent.

Explanation

Correct Answer

A:

A controlled comparison isolates the contribution of each change. Keeping the same evaluation cases also makes regressions visible instead of allowing them to disappear through selective testing.

Incorrect Answers

B:

Changing the test inputs introduces another variable. Differences might reflect the difficulty of the examples rather than the effect of the configuration.

C:

Best-case outputs hide failures and encourage selection bias. Diagnosis requires consistent measurement over the relevant examples, including the cases that became worse.

D:

Simultaneous changes confound the comparison. Even if the next run improves, the team still cannot tell which change helped or which introduced regressions.

Question 9

A Claude Project contains two expense policies with different meal limits. The newer document is approved; the older one is obsolete but remains in project knowledge. Claude sometimes cites the old limit. What is the best first correction?

  1. Ask for a friendlier tone when describing expenses.
  2. Remove or clearly segregate the obsolete policy and identify the approved current source.
  3. Repeat both limits in every prompt without explaining their status.
  4. Rename the project without changing its sources.

Explanation

Correct Answer

B:

Curating the knowledge base addresses the source conflict rather than treating each answer as an isolated failure. The current policy should be unambiguous and its approval status traceable.

Incorrect Answers

A:

Tone cannot resolve conflicting policy evidence. The response may become easier to read while continuing to apply an outdated limit.

C:

Presenting both values without precedence recreates the ambiguity. The model needs an authoritative basis for selecting the applicable policy.

D:

A new workspace name does not remove obsolete material. Correcting source versioning is the step that addresses why the wrong limit is being retrieved or used.

Question 10

Claude must classify requests as “billing,” “technical,” or “other.” It confuses billing questions that mention software errors with genuine technical incidents. Which addition best teaches the intended boundary?

  1. A request to make the category names more creative.
  2. An instruction to return all three labels for each request.
  3. More background about the company without any labeling examples.
  4. Examples of ambiguous requests with their approved labels and brief explanations of the distinction.

Explanation

Correct Answer

D:

Boundary examples demonstrate how the business rule applies when surface words are misleading. They should include both sides of the confusion so the model learns the decision criterion rather than a keyword shortcut.

Incorrect Answers

A:

Creative category names do not clarify the underlying rule. They may also break downstream processing that expects the three approved labels.

B:

Returning every label avoids making the required classification. It changes the task rather than improving the distinction between the existing categories.

C:

General context can help interpretation, but it does not directly resolve the observed ambiguity. Targeted labeled examples are more informative for this specific failure.

Claude Certified Associate - Foundations (CCAO-F) Practice Tests