Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 23

Question 220

A legal query has both a citation code and a prose concern, but search matches only one component. Which acceptance result provides the most specific evidence?

  1. A production-like test shows that multi-hop test cases preserve entity identities across each relation and cited document.
  2. The release check confirms that balance responses match the ledger after a simulated transaction update.
  3. The reviewer observes that paraphrased benchmark queries retrieve relevant repair procedures despite vocabulary differences.
  4. The scenario test demonstrates that evaluation covers both citation-ID accuracy and semantic question coverage.

Explanation

Correct Answer

D:

Hybrid search preserves exact statutory references while still matching paraphrased context; this checks the situation in which compliance questions combine a statute number with a natural-language risk description.

Incorrect Answers

A:

Relationships and entity constraints can be lost when treated as independent similarity matches; this applies when a researcher asks for multi-hop relationships across linked supplier records, but here a legal query has both a citation code and a prose concern, but search matches only one component.

B:

Fast-changing structured state requires live, authorized lookup rather than static document similarity; the alternative helps if an assistant answers 'What is the current account balance?' from a monthly indexed snapshot, instead of when a legal query has both a citation code and a prose concern, but search matches only one component.

C:

Embedding-based similarity can connect paraphrased symptoms to documentation that shares meaning not wording; it is justified if a support engineer describes symptoms in different words from the troubleshooting manual, not the case in which a legal query has both a citation code and a prose concern, but search matches only one component.

Question 221

The indexed billing corpus is refreshed monthly while a customer expects a minute-old transaction balance. Which acceptance check evaluates this requirement rather than a proxy?

  1. The observed outcome is that trend reports represent key categories across the full selected period rather than a few high-scoring chunks.
  2. The regression suite verifies that paraphrased benchmark queries retrieve relevant repair procedures despite vocabulary differences.
  3. A production-like test shows that balance responses match the ledger after a simulated transaction update.
  4. The release check confirms that every generated claim is supported by a permitted signed source in the acceptance suite.

Explanation

Correct Answer

C:

Fast-changing structured state requires live, authorized lookup rather than static document similarity; it is the decisive test when an assistant answers 'What is the current account balance?' from a monthly indexed snapshot.

Incorrect Answers

A:

Top-k retrieval optimized for point answers can underrepresent the corpus in global trend analysis; the other case is when a query asks for broad document trends rather than a single matching passage, not the present case where the indexed billing corpus is refreshed monthly while a customer expects a minute-old transaction balance.

B:

Embedding-based similarity can connect paraphrased symptoms to documentation that shares meaning not wording; this makes sense when a support engineer describes symptoms in different words from the troubleshooting manual, while this problem involves the indexed billing corpus is refreshed monthly while a customer expects a minute-old transaction balance.

D:

Authority-constrained retrieval is essential when a semantically relevant unofficial page is inadmissible; the trigger would be that a customer wants answers only from a small set of signed policy documents, rather than the risk that the indexed billing corpus is refreshed monthly while a customer expects a minute-old transaction balance.

Question 222

An investigator follows an ownership chain across suppliers, but a single-passage search misses the relationship. Which measurement is the most relevant release criterion?

  1. The scenario test demonstrates that balance responses match the ledger after a simulated transaction update.
  2. The end-to-end check confirms that evaluation covers both citation-ID accuracy and semantic question coverage.
  3. The recorded test establishes that trend reports represent key categories across the full selected period rather than a few high-scoring chunks.
  4. The regression suite verifies that multi-hop test cases preserve entity identities across each relation and cited document.
  5. A production-like test shows that every generated claim is supported by a permitted signed source in the acceptance suite.

Explanation

Correct Answer

D:

Relationships and entity constraints can be lost when treated as independent similarity matches; an acceptance test needs to cover cases where a researcher asks for multi-hop relationships across linked supplier records.

Incorrect Answers

A:

Fast-changing structured state requires live, authorized lookup rather than static document similarity; that is defensible if an assistant answers 'What is the current account balance?' from a monthly indexed snapshot, whereas the current issue is that an investigator follows an ownership chain across suppliers, but a single-passage search misses the relationship.

B:

Hybrid search preserves exact statutory references while still matching paraphrased context; the other workload is one where compliance questions combine a statute number with a natural-language risk description, whereas the release criterion addresses an investigator follows an ownership chain across suppliers, but a single-passage search misses the relationship.

C:

Top-k retrieval optimized for point answers can underrepresent the corpus in global trend analysis; this applies when a query asks for broad document trends rather than a single matching passage, while the required safeguard concerns an investigator follows an ownership chain across suppliers, but a single-passage search misses the relationship.

E:

Authority-constrained retrieval is essential when a semantically relevant unofficial page is inadmissible; it is justified if a customer wants answers only from a small set of signed policy documents, not this situation where an investigator follows an ownership chain across suppliers, but a single-passage search misses the relationship.

Question 223

A support assistant returns accurate answers but p95 latency spikes during complex reranking. Which implementation decision is the strongest fit?

  1. For the next architecture revision, set an evidence token budget and choose the highest-value passages using measured recall.
  2. For the current integration, load-test index concurrency and retrieval quality together, not latency alone.
  3. Under the system plan, route retrieval depth by query risk and ambiguity with predefined quality gates.
  4. As the design priority, profile retrieval stages and apply selective reranking only where it adds measurable value.

Explanation

Correct Answer

D:

Targeted reranking preserves difficult-query accuracy without imposing the same cost on every request; the audit trail can establish that a trace benchmark shows improved tail latency with bounded quality degradation.

Incorrect Answers

A:

Context pruning can reduce inference latency and cost when it preserves answer-bearing evidence; it is relevant if an answer generator receives many low-ranked chunks and uses most of its time parsing them, instead of when a support assistant returns accurate answers but p95 latency spikes during complex reranking.

B:

Search infrastructure pressure may damage ranking or timeout behavior before outright availability fails; it serves conditions where a RAG system scales to more simultaneous users and begins returning fewer relevant chunks, not the case in which a support assistant returns accurate answers but p95 latency spikes during complex reranking.

C:

Risk-adaptive search applies expensive verification only when correctness stakes justify the delay; the other case is when a critical legal workflow benefits from reranking but casual FAQ answers do not, whereas the problem presented is that a support assistant returns accurate answers but p95 latency spikes during complex reranking.

Question 224

A business query is served from a stale cache after a policy update. What should be prioritized during architecture review?

  1. In the project scope, route retrieval depth by query risk and ambiguity with predefined quality gates.
  2. Within the rollout design, profile retrieval stages and apply selective reranking only where it adds measurable value.
  3. For the next architecture revision, instrument separate timing for embedding, retrieval, reranking, inference and tool calls.
  4. For the current integration, tie cache invalidation to authoritative document versions and freshness requirements.

Explanation

Correct Answer

D:

Caching reduces latency only when its staleness window matches the business obligation; a pilot is successful when after a policy replacement, subsequent requests cite the new version inside the required interval.

Incorrect Answers

A:

Risk-adaptive search applies expensive verification only when correctness stakes justify the delay; that would be useful if a critical legal workflow benefits from reranking but casual FAQ answers do not, while this problem involves a business query is served from a stale cache after a policy update.

B:

Targeted reranking preserves difficult-query accuracy without imposing the same cost on every request; it serves a setting where a support assistant returns accurate answers but p95 latency spikes during complex reranking, and not where a business query is served from a stale cache after a policy update.

C:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; that is defensible if a service records slow end-to-end responses but cannot tell whether search or generation dominates, rather than the risk that a business query is served from a stale cache after a policy update.

Question 225

An answer generator receives many low-ranked chunks and uses most of its time parsing them. Which response follows from the actual operational constraint?

  1. For this workflow, set an evidence token budget and choose the highest-value passages using measured recall.
  2. For the affected service, tie cache invalidation to authoritative document versions and freshness requirements.
  3. In the project scope, scope cache keys to identity and entitlements and avoid caching sensitive volatile outputs indiscriminately.
  4. For the intended users, profile retrieval stages and apply selective reranking only where it adds measurable value.

Explanation

Correct Answer

A:

Context pruning can reduce inference latency and cost when it preserves answer-bearing evidence; one measurable outcome is that budgeted retrieval meets answer recall targets while reducing input tokens per query.

Incorrect Answers

B:

Caching reduces latency only when its staleness window matches the business obligation; that belongs to situations where a business query is served from a stale cache after a policy update, while the required safeguard concerns an answer generator receives many low-ranked chunks and uses most of its time parsing them.

C:

Latency optimization must not bypass authorization or freshness boundaries; it is relevant if an engineer proposes caching all user responses even when access changes frequently, not the risk where an answer generator receives many low-ranked chunks and uses most of its time parsing them.

D:

Targeted reranking preserves difficult-query accuracy without imposing the same cost on every request; it serves conditions where a support assistant returns accurate answers but p95 latency spikes during complex reranking, not this situation where an answer generator receives many low-ranked chunks and uses most of its time parsing them.

Question 226

A critical legal workflow benefits from reranking but casual FAQ answers do not. Which change targets the constraint rather than a proxy metric?

  1. At the system boundary, tie cache invalidation to authoritative document versions and freshness requirements.
  2. For the primary workstream, instrument separate timing for embedding, retrieval, reranking, inference and tool calls.
  3. For this workflow, route retrieval depth by query risk and ambiguity with predefined quality gates.
  4. For the affected service, set an evidence token budget and choose the highest-value passages using measured recall.
  5. In the project scope, load-test index concurrency and retrieval quality together, not latency alone.

Explanation

Correct Answer

C:

Risk-adaptive search applies expensive verification only when correctness stakes justify the delay; the audit trail can establish that high-stakes cases retain citation quality while routine cases satisfy their response SLA.

Incorrect Answers

A:

Caching reduces latency only when its staleness window matches the business obligation; that fits cases where a business query is served from a stale cache after a policy update, not the case in which a critical legal workflow benefits from reranking but casual FAQ answers do not.

B:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; this control is needed when a service records slow end-to-end responses but cannot tell whether search or generation dominates, whereas the problem presented is that a critical legal workflow benefits from reranking but casual FAQ answers do not.

D:

Context pruning can reduce inference latency and cost when it preserves answer-bearing evidence; it serves a setting where an answer generator receives many low-ranked chunks and uses most of its time parsing them, but the actual risk is that a critical legal workflow benefits from reranking but casual FAQ answers do not.

E:

Search infrastructure pressure may damage ranking or timeout behavior before outright availability fails; that is defensible if a RAG system scales to more simultaneous users and begins returning fewer relevant chunks, but the requested boundary concerns a critical legal workflow benefits from reranking but casual FAQ answers do not.

Question 227

A service records slow end-to-end responses but cannot tell whether search or generation dominates. Which action is justified by the scenario?

  1. For the intended users, scope cache keys to identity and entitlements and avoid caching sensitive volatile outputs indiscriminately.
  2. Within the delivery plan, route retrieval depth by query risk and ambiguity with predefined quality gates.
  3. At the system boundary, instrument separate timing for embedding, retrieval, reranking, inference and tool calls.
  4. For the primary workstream, tie cache invalidation to authoritative document versions and freshness requirements.

Explanation

Correct Answer

C:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; a pilot is successful when a sample trace attributes elapsed time to each stage and identifies the dominant component.

Incorrect Answers

A:

Latency optimization must not bypass authorization or freshness boundaries; its scope includes cases where an engineer proposes caching all user responses even when access changes frequently, and not where a service records slow end-to-end responses but cannot tell whether search or generation dominates.

B:

Risk-adaptive search applies expensive verification only when correctness stakes justify the delay; it targets cases where a critical legal workflow benefits from reranking but casual FAQ answers do not, rather than the risk that a service records slow end-to-end responses but cannot tell whether search or generation dominates.

D:

Caching reduces latency only when its staleness window matches the business obligation; that belongs to situations where a business query is served from a stale cache after a policy update, while the immediate question involves a service records slow end-to-end responses but cannot tell whether search or generation dominates.

Question 228

An engineer proposes caching all user responses even when access changes frequently. What change directly resolves this production-design issue?

  1. Under the reviewed approach, instrument separate timing for embedding, retrieval, reranking, inference and tool calls.
  2. For this production path, scope cache keys to identity and entitlements and avoid caching sensitive volatile outputs indiscriminately.
  3. For the intended users, tie cache invalidation to authoritative document versions and freshness requirements.
  4. Within the delivery plan, load-test index concurrency and retrieval quality together, not latency alone.

Explanation

Correct Answer

B:

Latency optimization must not bypass authorization or freshness boundaries; release assurance requires proof that role-change probes cannot read a prior user's cached result.

Incorrect Answers

A:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; the trigger would be that a service records slow end-to-end responses but cannot tell whether search or generation dominates, while the required safeguard concerns an engineer proposes caching all user responses even when access changes frequently.

C:

Caching reduces latency only when its staleness window matches the business obligation; that fits cases where a business query is served from a stale cache after a policy update, not this situation where an engineer proposes caching all user responses even when access changes frequently.

D:

Search infrastructure pressure may damage ranking or timeout behavior before outright availability fails; this control is needed when a RAG system scales to more simultaneous users and begins returning fewer relevant chunks, whereas this scenario involves an engineer proposes caching all user responses even when access changes frequently.

Question 229

A RAG system scales to more simultaneous users and begins returning fewer relevant chunks. Which proposed design action directly addresses the risk?

  1. In the initial solution, profile retrieval stages and apply selective reranking only where it adds measurable value.
  2. In the proposed design, instrument separate timing for embedding, retrieval, reranking, inference and tool calls.
  3. Under the reviewed approach, load-test index concurrency and retrieval quality together, not latency alone.
  4. During architecture review, scope cache keys to identity and entitlements and avoid caching sensitive volatile outputs indiscriminately.

Explanation

Correct Answer

C:

Search infrastructure pressure may damage ranking or timeout behavior before outright availability fails; a representative trial should prove that peak-load trials meet both retrieval recall and request-time targets.

Incorrect Answers

A:

Targeted reranking preserves difficult-query accuracy without imposing the same cost on every request; the alternative helps if a support assistant returns accurate answers but p95 latency spikes during complex reranking, whereas the problem presented is that a RAG system scales to more simultaneous users and begins returning fewer relevant chunks.

B:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; it is justified if a service records slow end-to-end responses but cannot tell whether search or generation dominates, but this decision concerns a RAG system scales to more simultaneous users and begins returning fewer relevant chunks.

D:

Latency optimization must not bypass authorization or freshness boundaries; it targets cases where an engineer proposes caching all user responses even when access changes frequently, but the requested boundary concerns a RAG system scales to more simultaneous users and begins returning fewer relevant chunks.

Question 230

At peak time, a ranking stage adds large tail delays despite generating accurate source-grounded answers. Which verified outcome most directly addresses the reported risk?

  1. A reproducible run shows that role-change probes cannot read a prior user's cached result.
  2. The signed test record confirms that a trace benchmark shows improved tail latency with bounded quality degradation.
  3. The sign-off record demonstrates that a sample trace attributes elapsed time to each stage and identifies the dominant component.
  4. The acceptance evidence shows that budgeted retrieval meets answer recall targets while reducing input tokens per query.

Explanation

Correct Answer

B:

Targeted reranking preserves difficult-query accuracy without imposing the same cost on every request; the test is necessary where a support assistant returns accurate answers but p95 latency spikes during complex reranking.

Incorrect Answers

A:

Latency optimization must not bypass authorization or freshness boundaries; this makes sense when an engineer proposes caching all user responses even when access changes frequently, rather than the risk that at peak time, a ranking stage adds large tail delays despite generating accurate source-grounded answers.

C:

Stage-level telemetry directs optimization to the actual bottleneck rather than speculative tuning; the trigger would be that a service records slow end-to-end responses but cannot tell whether search or generation dominates, while the immediate question involves at peak time, a ranking stage adds large tail delays despite generating accurate source-grounded answers.

D:

Context pruning can reduce inference latency and cost when it preserves answer-bearing evidence; the design helps when an answer generator receives many low-ranked chunks and uses most of its time parsing them, but the present constraint is that at peak time, a ranking stage adds large tail delays despite generating accurate source-grounded answers.

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