Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 15

Question 140

A reusable procedure explains how to reconcile invoices but embeds a supplier bank account that changes occasionally. What separation is best?

  1. Keep the procedure reusable and resolve bank details from approved live data
  2. Copy the account into more examples to reinforce it
  3. Treat the embedded account as permanent because the procedure is approved
  4. Reuse the bank details embedded in the last successful execution

Explanation

Correct Answer

A:

Procedural knowledge and mutable operational facts have different lifecycles. The workflow should retrieve the current authorized account rather than trust an old embedded value.

Incorrect Answers

B:

Repetition does not improve freshness or authority. It makes the stale detail more influential while preserving the underlying update problem.

C:

Approval of a procedure does not make every included fact timeless. Mutable bank details need their own verification and update path.

D:

A previous successful payment does not establish current account details. The procedure should resolve authoritative live values for the new transaction.

Question 141

An application inserts user text into a prompt template using the same markers that delimit trusted sections. What should the template system ensure?

  1. Every user message is inserted as a new developer instruction
  2. User content cannot alter the template's structural boundaries
  3. Longer markers alone guarantee complete injection resistance
  4. User input is allowed to redefine the section markers

Explanation

Correct Answer

B:

Safe interpolation preserves the distinction between supplied data and application instructions. Delimiter handling should prevent user text from closing or replacing trusted sections.

Incorrect Answers

A:

Promoting input to a trusted role changes its authority. The application should preserve the intended data role and context boundaries.

C:

Distinct markers can reduce accidental collisions but are not a complete security boundary. Structural handling and downstream controls are still necessary.

D:

That gives untrusted content control over the template's structure. It undermines the separation the markers were intended to provide.

Question 142

The same policy paragraph appears in the system prompt, examples, retrieved context, and conversation summary. What should a context audit examine?

  1. Whether repeating the copies in a larger window reduces their relative proportion
  2. Whether the final answer can omit the policy entirely
  3. Whether all unrelated evidence can be deleted to fit the copies
  4. Whether redundant copies add value or introduce inconsistent versions

Explanation

Correct Answer

D:

Duplication consumes space and can create conflicts when copies drift. The audit should preserve necessary guidance while removing redundant or contradictory representations.

Incorrect Answers

A:

A larger window changes available capacity but does not establish that duplicate evidence is useful or consistent. The redundant copies themselves need examination.

B:

Output content does not resolve duplicated input representations. The problem should be addressed where the context is assembled.

C:

Removing relevant task evidence to preserve redundant policy text reverses the priority. The goal is useful context, not maximal repetition.

Question 143

A prompt experiment simultaneously changes the role, examples, output schema, and evidence order. Results improve, but the cause is unclear. What should follow?

  1. Attribute the improvement to the first edited component
  2. Keep adding changes until the explanation feels plausible
  3. Assume every changed component improved performance equally
  4. Ablate components to identify which changes contribute

Explanation

Correct Answer

D:

Controlled removal or substitution can separate the effects of prompt components. This helps retain useful changes without preserving unnecessary complexity.

Incorrect Answers

A:

Edit order is not causal evidence. Several changes occurred together, so their individual contributions remain unresolved.

B:

Additional simultaneous changes make attribution harder. The next experiment should reduce uncertainty, not increase the number of confounded factors.

C:

Equal contribution is unsupported. Some changes may help, others may be neutral, and some may harm cases hidden by the aggregate result.

Question 144

A system prompt calls Claude a world-leading auditor. The team claims this establishes factual reliability. Which assessment is accurate?

  1. The role title can substitute for source checks when the response sounds professional
  2. The title proves the model has been evaluated on this organization's audit criteria
  3. Role framing can guide style but does not verify facts
  4. Agreement between two answers using the same persona establishes factual accuracy

Explanation

Correct Answer

C:

A persona may shape tone and task focus. Reliability still depends on evidence, evaluation, and checks appropriate to the actual auditing decisions.

Incorrect Answers

A:

A professional tone is not evidence that the underlying statements are correct. Factual claims still require suitable evidence and validation.

B:

A prompt persona does not establish the content or results of an evaluation. Organization-specific criteria need to be tested directly.

D:

Two outputs can share the same unsupported assumption. Agreement under a role prompt does not independently verify either answer against the facts.

Question 145

An assistant uses phrases such as definitely and guaranteed even when evidence is incomplete. Which prompt and evaluation change is most appropriate?

  1. Replace every assertion with maybe regardless of support
  2. Remove all uncertainty language to keep responses decisive
  3. Ask for a confidence percentage without testing its meaning
  4. Define evidence-dependent uncertainty behavior and test it

Explanation

Correct Answer

D:

The task should specify when to qualify, abstain, or request information. Evaluation must check whether those expressions track evidence rather than merely sound cautious.

Incorrect Answers

A:

Universal hedging obscures well-supported facts and does not calibrate uncertainty. The language should reflect the strength and completeness of evidence.

B:

Decisiveness can conceal missing information. The requirement is reliable communication of evidence limits, not uniform rhetorical confidence.

C:

A number can appear precise without being calibrated. Its relationship to actual error rates needs evidence before users rely on it.

Question 146

Tool schemas contain long marketing paragraphs but omit parameter units and accepted ranges. Which edit is most valuable?

  1. Abbreviate all parameter names to single letters
  2. Move units and ranges into a separate document unavailable during tool selection
  3. Replace promotional prose with operational parameter semantics
  4. Add more successful call examples while leaving parameter constraints implicit

Explanation

Correct Answer

C:

Selection and execution depend on what inputs mean and what values are valid. Concise functional descriptions improve usefulness per token.

Incorrect Answers

A:

Short names can increase ambiguity and argument errors. Token reduction should preserve or improve the clarity of the interface.

B:

Hidden operational semantics cannot help the model construct valid arguments. The callable interface needs accessible descriptions of the parameters it accepts.

D:

Examples illustrate particular calls but may not reveal the full accepted range or units. Explicit parameter semantics resolve the stated schema gap more directly.

Question 147

An English prompt works well, but the service must answer Spanish requests using the same policy. What should be tested before launch?

  1. Policy fidelity and response quality on representative Spanish cases
  2. Use English test scores as a complete substitute
  3. Only whether the output contains Spanish words
  4. Translate one easy example and declare full coverage

Explanation

Correct Answer

A:

Language changes can affect interpretation and terminology. The shared policy goal needs evidence on the target language, not an assumption of perfect transfer.

Incorrect Answers

B:

English performance is relevant background but does not prove target-language behavior. The deployed input distribution must be represented in testing.

C:

Language identification does not establish policy correctness, completeness, or useful terminology. A fluent Spanish response can still apply the wrong rule.

D:

A single case does not represent the variety or risks of production requests. The evaluation should cover meaningful decision boundaries in Spanish.

Question 148

A request combines system instructions, tool schemas, retrieved passages, and a conversation summary. Which checks should occur before submission? Choose THREE.

  1. Verify that required instructions and active constraints are present
  2. Assume every retrieved passage is authoritative instruction
  3. Treat a successful previous request as proof this one fits
  4. Confirm that supplied evidence is authorized for this request
  5. Check the assembled token budget against supported limits
  6. Drop source identifiers whenever the context is large

Explanation

Correct Answers

A:

A request can fit technically while omitting the rules governing the task. Required instructions must survive filtering and compaction.

D:

Context assembly is a disclosure boundary. Relevant information still must be permitted for the requesting user and workflow.

E:

Individual components may fit separately but overflow when combined. Count the actual assembled request and reserve the needed response capacity.

Incorrect Answers

B:

Retrieved material is evidence, not automatic governing authority. Its role and trust level must remain distinct from application instructions.

C:

Inputs and retrieved material can change across requests. A prior success does not establish the budget or authorization of the current assembly.

F:

Identifiers may be necessary for verification and provenance. Budget pressure should not silently remove information required by the task contract.

Question 149

A delivery tool expects weight in kilograms, but its schema says only weight: number. Calls sometimes contain pounds. What interface change best addresses the root cause?

  1. Rename the tool without changing its parameters
  2. Increase the accepted numeric range without specifying a unit
  3. Make the unit explicit and validate the accepted range
  4. Convert every number by a fixed factor unconditionally

Explanation

Correct Answer

C:

The interface must communicate the meaning of the numeric input, not just its type. Validation can reject implausible values before execution.

Incorrect Answers

A:

A different function name may still leave the same unit ambiguity. The parameter semantics need to be explicit where arguments are formed.

B:

A broader numeric range does not distinguish pounds from kilograms. The schema must specify the weight unit so argument construction and validation use the same interpretation.

D:

Unconditional conversion corrupts inputs already expressed in kilograms. Conversion requires a known source unit rather than an assumption about every call.

Question 150

Two tools have similar names, but one previews a customer change and the other commits it. How should their interfaces distinguish them?

  1. Expose the preview operation first and leave both descriptions unchanged
  2. State effects and required authority clearly in each description
  3. Give both tools the same description for consistency
  4. Add detailed output examples without stating which operation writes
  5. Hide the preview tool so the model has fewer choices

Explanation

Correct Answer

B:

Tool selection depends on understanding consequences. Explicit preview-versus-commit semantics reduce accidental use while execution-side authorization enforces the actual boundary.

Incorrect Answers

A:

Ordering may influence selection but does not explain which operation commits a change. The tool contract must communicate effects and required authority.

C:

Identical descriptions hide the important behavioral difference. Consistent formatting should not erase the distinction between observation and modification.

D:

Examples can clarify result shape while leaving side effects ambiguous. The decisive distinction is whether invocation changes external state.

E:

Reducing choices by removing the safe preparation capability can make the workflow less controllable. The issue is clarity, not simply tool count.

Claude Certified Architect - Professional (CCAR-P) Practice Tests