Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 11

Question 100

A long customer case exceeds the model's context budget after attaching years of chat history. Which next step is most defensible?

  1. Maintain a structured issue state and expire low-value transcripts under retention policy.
  2. Summarize resolved history into verified state and keep only relevant recent turns and evidence.
  3. Send role-specific task context and compact intermediate artifacts with trace references.
  4. Use section-aware indexing and fetch the relevant clauses plus nearby definitions on demand.

Explanation

Correct Answer

B:

A compact durable state preserves decision-critical facts without repeatedly transmitting stale conversation. Preserve the evidence required for relevant-context compression inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

This shifts attention toward structured session memory when the relevant decision is about relevant-context compression. Unless the architecture can summarize resolved history into verified state and keep only relevant recent turns and evidence, the failure described in the stem remains unresolved.

C:

Although minimal delegated context can be valuable, it is not the acceptance condition in the question. The missing provision is relevant-context compression, addressed by the control to summarize resolved history into verified state and keep only relevant recent turns and evidence.

D:

It is aimed at a different outcome, hierarchical retrieval for large documents. No evidence is given that it provides relevant-context compression; the needed mechanism is to summarize resolved history into verified state and keep only relevant recent turns and evidence at the appropriate boundary.

Question 101

A documentation bot includes the entire product manual when each query touches two paragraphs. Which approach should be recommended?

  1. Use section-aware indexing and fetch the relevant clauses plus nearby definitions on demand.
  2. Preserve table structure and headers when constructing evidence chunks.
  3. Retrieve a small, permission-filtered set of relevant chunks instead of bulk-inserting documents.
  4. Summarize resolved history into verified state and keep only relevant recent turns and evidence.

Explanation

Correct Answer

C:

Selective retrieval controls input tokens and reduces distracting material in the model's attention. Preserve the evidence required for targeted context selection inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

There is a legitimate use case for hierarchical retrieval for large documents, but it does not supply the test or enforcement needed for targeted context selection. The design needs to retrieve a small, permission-filtered set of relevant chunks instead of bulk-inserting documents.

B:

Consider its actual operating scope: structure-preserving context engineering. The requested assurance is targeted context selection, which calls for an implementation that can retrieve a small, permission-filtered set of relevant chunks instead of bulk-inserting documents, not merely for a related improvement.

D:

The proposed measure is aimed at relevant-context compression, not targeted context selection. The architecture must instead retrieve a small, permission-filtered set of relevant chunks instead of bulk-inserting documents; a separate change is still needed for targeted context selection.

Question 102

An agent repeats the same large tool descriptions on every call despite using only three operations. What is the best production design response?

  1. Deduplicate retrieval results and use diversification or reranking before prompt assembly.
  2. Use section-aware indexing and fetch the relevant clauses plus nearby definitions on demand.
  3. Expose essential tools and use deferred discovery for rarely needed capabilities where supported.
  4. Send role-specific task context and compact intermediate artifacts with trace references.

Explanation

Correct Answer

C:

Reducing always-loaded schemas limits context overhead and accidental tool selection. Preserve the evidence required for scoped tool context inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

By itself, this improves nonredundant retrieval context, but the case is not asking the architect to optimize that dimension. The requirement is scoped tool context, supported by an approach to expose essential tools and use deferred discovery for rarely needed capabilities where supported.

B:

Its main effect concerns hierarchical retrieval for large documents. The case explicitly hinges on scoped tool context; to resolve it reliably, the implementation has to expose essential tools and use deferred discovery for rarely needed capabilities where supported, which this choice does not establish.

D:

The proposed action prioritizes minimal delegated context. That can coexist with the right solution, but it cannot replace scoped tool context: the application must still expose essential tools and use deferred discovery for rarely needed capabilities where supported.

Question 103

A summarizer cuts off the final compliance section because output tokens are exhausted. Which change most directly satisfies this requirement?

  1. Preserve table structure and headers when constructing evidence chunks.
  2. Reserve an output budget and explicitly structure priorities before generation.
  3. Use section-aware indexing and fetch the relevant clauses plus nearby definitions on demand.
  4. Maintain a structured issue state and expire low-value transcripts under retention policy.

Explanation

Correct Answer

B:

Context capacity must account for the response as well as inputs; blindly maximizing retrieved text can crowd out completion.

Incorrect Answers

A:

It is aimed at a different outcome, structure-preserving context engineering. No evidence is given that it provides output-budget-aware context planning; the needed mechanism is to reserve an output budget and explicitly structure priorities before generation at the appropriate boundary.

C:

That is primarily a technique for hierarchical retrieval for large documents. It lacks the explicit measure for output-budget-aware context planning that is necessary here, namely a system that will reserve an output budget and explicitly structure priorities before generation.

D:

Improving structured session memory does not imply output-budget-aware context planning. The selected design must reserve an output budget and explicitly structure priorities before generation, whereas this choice omits that direct protection or verification step.

Question 104

A search tool returns hundreds of redundant paragraphs from near-identical policy documents. Which solution best resolves this operational gap?

  1. Preserve table structure and headers when constructing evidence chunks.
  2. Deduplicate retrieval results and use diversification or reranking before prompt assembly.
  3. Expose essential tools and use deferred discovery for rarely needed capabilities where supported.
  4. Rerank for relevance and place authoritative instructions and supporting evidence clearly.

Explanation

Correct Answer

B:

Removing overlaps increases evidence coverage per token and avoids overweighting duplicated sources. Preserve the evidence required for nonredundant retrieval context inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

The proposed measure is aimed at structure-preserving context engineering, not nonredundant retrieval context. The architecture must instead deduplicate retrieval results and use diversification or reranking before prompt assembly; a separate change is still needed for nonredundant retrieval context.

C:

The control would be justified if scoped tool context were the limiting factor. In this scenario the decisive target is nonredundant retrieval context; success requires an architecture that will deduplicate retrieval results and use diversification or reranking before prompt assembly.

D:

The option belongs to the broader work on salient evidence placement, but it cannot by itself deliver nonredundant retrieval context. A decision on this case needs the team to deduplicate retrieval results and use diversification or reranking before prompt assembly and verify that outcome.

Question 105

A support system provides outdated account balances in chat history even after a fresh balance tool result is returned. What design decision would best fit these facts?

  1. Rerank for relevance and place authoritative instructions and supporting evidence clearly.
  2. Summarize resolved history into verified state and keep only relevant recent turns and evidence.
  3. Expose essential tools and use deferred discovery for rarely needed capabilities where supported.
  4. Preserve table structure and headers when constructing evidence chunks.
  5. Prioritize authoritative current tool data and label timestamps over older conversational claims.

Explanation

Correct Answer

E:

Freshness and provenance should determine which facts survive context construction. Preserve the evidence required for fresh authoritative fact precedence inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

The proposed action prioritizes salient evidence placement. That can coexist with the right solution, but it cannot replace fresh authoritative fact precedence: the application must still prioritize authoritative current tool data and label timestamps over older conversational claims.

B:

The design concern is real but different: relevant-context compression. It does not impose the specific requirement for fresh authoritative fact precedence; a suitable solution must prioritize authoritative current tool data and label timestamps over older conversational claims and demonstrate the result.

C:

This addresses scoped tool context; however, the problem asks for fresh authoritative fact precedence. Without the safeguard to prioritize authoritative current tool data and label timestamps over older conversational claims, the original failure mode is still possible.

D:

That choice would be responsive to structure-preserving context engineering, but the reported defect is different. It leaves fresh authoritative fact precedence unproved because there is no commitment to prioritize authoritative current tool data and label timestamps over older conversational claims.

Question 106

Three specialist agents investigate a time-ordered outage. A compression step summarizes logs but strips time zones, event identifiers and units, making the workers disagree about what happened first. What context-engineering change best protects the diagnosis?

  1. Replace exact times with words such as early and late to reduce prompt tokens even further.
  2. Send every investigator all unfiltered raw logs on every model step to eliminate summaries altogether.
  3. Choose whichever worker sounds most confident and publish its incident timeline.
  4. Share a compact structured event timeline preserving source IDs, timestamp zones and units, with trace links to the underlying records.

Explanation

Correct Answer

D:

Compression is useful only if it preserves fields essential to the inference. A typed timeline with source provenance lets specialists compare chronological evidence without reproducing huge raw logs or silently changing meanings.

Incorrect Answers

A:

Approximate ordering destroys the time resolution needed to establish causality and can reverse event sequences across time zones. Token savings cannot justify dropping the disputed fields.

B:

Unbounded logs create context and cost problems and may still bury critical events. Compact verified event fields and direct trace references preserve auditability more effectively.

C:

Linguistic confidence is not a substitute for reconciled timestamps and event identities. The disputed chronology must be checked against authoritative records.

Question 107

An assistant gives erratic answers when the critical policy paragraph is buried among dozens of irrelevant excerpts. Which design choice best matches the stated constraint?

  1. Rerank for relevance and place authoritative instructions and supporting evidence clearly.
  2. Prioritize authoritative current tool data and label timestamps over older conversational claims.
  3. Deduplicate retrieval results and use diversification or reranking before prompt assembly.
  4. Preserve table structure and headers when constructing evidence chunks.

Explanation

Correct Answer

A:

Context quality and positioning matter; more documents are not a substitute for selecting the right ones. Preserve the evidence required for salient evidence placement inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

B:

This shifts attention toward fresh authoritative fact precedence when the relevant decision is about salient evidence placement. Unless the architecture can rerank for relevance and place authoritative instructions and supporting evidence clearly, the failure described in the stem remains unresolved.

C:

For a workflow whose primary weakness was nonredundant retrieval context, that proposal could be reasonable. Here the unmet obligation is salient evidence placement, and that requires the system to rerank for relevance and place authoritative instructions and supporting evidence clearly.

D:

Although structure-preserving context engineering can be valuable, it is not the acceptance condition in the question. The missing provision is salient evidence placement, addressed by the control to rerank for relevance and place authoritative instructions and supporting evidence clearly.

Question 108

A retrieval system splits tables into fragments without column labels, causing incorrect numerical conclusions. Which architecture adjustment is most appropriate?

  1. Rerank for relevance and place authoritative instructions and supporting evidence clearly.
  2. Reserve an output budget and explicitly structure priorities before generation.
  3. Deduplicate retrieval results and use diversification or reranking before prompt assembly.
  4. Preserve table structure and headers when constructing evidence chunks.

Explanation

Correct Answer

D:

Token reduction must not destroy relationships necessary for interpreting a retrieved cell. Preserve the evidence required for structure-preserving context engineering inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

That choice would be responsive to salient evidence placement, but the reported defect is different. It leaves structure-preserving context engineering unproved because there is no commitment to preserve table structure and headers when constructing evidence chunks.

B:

There is a legitimate use case for output-budget-aware context planning, but it does not supply the test or enforcement needed for structure-preserving context engineering. The design needs to preserve table structure and headers when constructing evidence chunks.

C:

Consider its actual operating scope: nonredundant retrieval context. The requested assurance is structure-preserving context engineering, which calls for an implementation that can preserve table structure and headers when constructing evidence chunks, not merely for a related improvement.

Question 109

An incident chat needs to remember unresolved actions but can discard raw debug output after resolution. Which next step is most defensible?

  1. Reserve an output budget and explicitly structure priorities before generation.
  2. Maintain a structured issue state and expire low-value transcripts under retention policy.
  3. Summarize resolved history into verified state and keep only relevant recent turns and evidence.
  4. Send role-specific task context and compact intermediate artifacts with trace references.

Explanation

Correct Answer

B:

Intentional memory scopes preserve actionable continuity while reducing long-context cost and exposure. Preserve the evidence required for structured session memory inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

This would make sense as an additional project activity for output-budget-aware context planning. The priority required here is structured session memory, so the workflow still needs a control that will maintain a structured issue state and expire low-value transcripts under retention policy.

C:

Its main effect concerns relevant-context compression. The case explicitly hinges on structured session memory; to resolve it reliably, the implementation has to maintain a structured issue state and expire low-value transcripts under retention policy, which this choice does not establish.

D:

The option might improve minimal delegated context; that remains insufficient for structured session memory. The case is resolved by structured session memory, which depends on a design that will maintain a structured issue state and expire low-value transcripts under retention policy.

Question 110

A contract assistant answers questions that depend on definitions in a master agreement and a revised liability limit in a separately signed addendum. Similarity search retrieves only the master agreement, so the controlling amendment is omitted. Which retrieval design is most suitable?

  1. Rely only on the highest-scoring passage from the master agreement because it has the strongest keyword similarity.
  2. Concatenate every historic agreement in the organization into the prompt and omit document-level provenance.
  3. Ask the model to guess likely liability limits from general commercial practice when the cited agreement is silent.
  4. Index cross-document relationships and effective dates, then retrieve the master clause and its controlling addendum together with citations.

Explanation

Correct Answer

D:

A current amendment can override otherwise relevant older text. Following explicit agreement-to-addendum links and checking signed effective versions prevents an apparently well-grounded answer from omitting a controlling instrument.

Incorrect Answers

A:

A similarity score cannot establish which contractual instrument governs a clause. The signed addendum can have lower lexical similarity while controlling the current liability limit.

B:

Bulk context makes controlling versions harder to identify and can exceed capacity. Contractual precedence and specific source attribution remain necessary even when all text is technically visible.

C:

Commercial conventions are not evidence of the parties’ signed amendment. The assistant should resolve related controlling instruments, not invent terms to fill retrieval gaps.

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