Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 5

Question 40

An assistant spends half its context replaying a month of old chat when only three decisions still matter. What is the best memory pattern?

  1. Keep a compact, explicit summary of durable decisions and retrieve details when needed.
  2. Keep only the most recent sentence and discard the three durable decisions.
  3. As an alternative, compute aggregates with a tool and pass only validated results back to Claude.
  4. Provide the web-reading worker limited tools and treat its findings as data, not instructions.

Explanation

Correct Answer

A:

Separating long-term task facts from raw transcript prevents context waste while preserving the information needed for new decisions.

Incorrect Answers

B:

A last-sentence-only summary loses the decisions the scenario explicitly still needs. The model may save tokens initially, but the next turn would lack essential task constraints and commit history.

C:

Moving mechanical data processing into code avoids token-heavy raw inputs while keeping Claude focused on interpretation; even if offloading deterministic computation succeeds, compact durable memory still needs separate verification.

D:

Isolating untrusted retrieval prevents document text from inheriting authority over high-impact actions in the parent process; for compact durable memory, a control centered on untrusted research containment is not the appropriate substitute.

Question 41

An agent must browse many documents and later cite specific statements. How should intermediate evidence be stored?

  1. Preserve source identifiers and selected excerpts in structured working memory.
  2. Store task status and identifiers in an external store rather than replaying every token.
  3. Execute the critic in a separate context with the draft and evaluation rubric.
  4. Send a constrained result object containing findings, confidence, and evidence IDs.

Explanation

Correct Answer

A:

Evidence provenance makes later claims checkable; keeping only an unreferenced summary would lose the link to the supporting records.

Incorrect Answers

B:

Durable state records progress across sessions and can be validated; full transcript replay is costly and may not reflect authoritative system state; changing cross-session task continuity does not establish provenance in agent memory in this application.

C:

Isolated critique reduces anchoring to exploratory reasoning and yields a more independent check of the artifact; for provenance in agent memory, this instead manages independent evaluator context; the requested decision remains unresolved.

D:

A structured contract limits what crosses the context boundary and provides information the coordinator can verify; the effect concerns structured specialist handoff; the success condition is provenance in agent memory.

Question 42

A multi-step workflow needs to resume after a database outage. Which state design helps?

  1. Keep source identifiers and selected excerpts in structured working memory.
  2. Ask the worker for a bounded evidence summary with references to the originals.
  3. Assess the execution trace and resulting system state against task criteria.
  4. Checkpoint the workflow position and completed side effects outside process memory.

Explanation

Correct Answer

D:

Durable checkpoints let the executor resume without blindly rerunning earlier writes after a crash. Persist the last committed transition and operation identifiers so a resumed worker can distinguish pending work from already completed writes.

Incorrect Answers

A:

Evidence provenance makes later claims checkable; keeping only an unreferenced summary would lose the link to the supporting records; this may help provenance in agent memory but cannot by itself deliver crash-safe workflow checkpoints.

B:

Compact outputs preserve relevant facts and sources while preventing the parent context from being overwhelmed by intermediate material; where crash-safe workflow checkpoints matters, a control for concise worker return contracts leaves that test unmet.

C:

Multi-step success depends on tool choices and side effects, not just fluent output; trajectory evidence reveals hidden failures; it changes state-based agent evaluation; it does not prove crash-safe workflow checkpoints in this case.

Question 43

A supervisor has a strict search quota shared by five worker agents. Urgent fraud cases must not be starved by low-priority background research. What scheduling policy should govern the shared tool?

  1. Let each worker issue unlimited search requests without coordinating with the others.
  2. Always serve the oldest request first without distinguishing urgent incident work.
  3. Use one quota-aware work queue with explicit priority and fairness rules across worker requests.
  4. Give urgent workers longer natural-language instructions instead of changing request admission.

Explanation

Correct Answer

C:

A shared scheduler can reserve capacity for urgent cases while preventing starvation of ordinary jobs. Central admission control makes the service quota and work priorities visible outside the model-generated plans.

Incorrect Answers

A:

Independent loops cannot see the aggregate quota or coordinate urgent work. They can exhaust the shared allowance and leave time-critical investigations waiting.

B:

Pure arrival order gives no protection to a time-critical case queued behind large background jobs. The stated priority requirement needs a controlled scheduling policy.

D:

Prompt verbosity does not reserve upstream search capacity or move work ahead in a dispatcher. The constraint exists in the shared execution service rather than in the wording of worker tasks.

Question 44

A tool-based agent occasionally cycles between search and read without new evidence. What loop control should detect this?

  1. Save stable configuration in versioned files and load it into each session.
  2. Employ durable state and idempotent tool operations across retries.
  3. Save task status and identifiers in an external store rather than replaying every token.
  4. Track repeated action-observation signatures and terminate or replan on repetition.

Explanation

Correct Answer

D:

Cycle detection identifies lack of progress independently of the model's confidence and prevents unbounded redundant calls.

Incorrect Answers

A:

Reproducible configuration should be external to ephemeral conversation memory, making changes reviewable and repeatable; for this request, repeated action loop detection is the missing piece, not versioned durable configuration.

B:

Recovery requires both knowledge of committed stages and protection from duplicate writes; neither a model summary nor blind retries supplies that guarantee; testing repeated action loop detection would still fail after improving side-effect safe recovery.

C:

Durable state records progress across sessions and can be validated; full transcript replay is costly and may not reflect authoritative system state; a remedy for cross-session task continuity is not a substitute for repeated action loop detection.

Question 45

An operations team wants a visible graph of branching tasks, retries, and approval checkpoints. What kind of abstraction helps?

  1. Save the workflow position and completed side effects outside process memory.
  2. Provide a constrained result object containing findings, confidence, and evidence IDs.
  3. As an alternative, demonstrate a needed capability such as durable state, branching, or observability.
  4. Use a workflow graph with explicit states and transition rules.

Explanation

Correct Answer

D:

Graph-based orchestration makes conditional edges and retry points inspectable, rather than hiding all sequencing in natural-language planning.

Incorrect Answers

A:

Durable checkpoints let the executor resume without blindly rerunning earlier writes after a crash; this may help crash-safe workflow checkpoints but cannot by itself deliver inspectable task-state graph.

B:

A structured contract limits what crosses the context boundary and provides information the coordinator can verify; testing inspectable task-state graph would still fail after improving structured specialist handoff.

C:

Framework overhead is worthwhile when its orchestration features solve a real problem, not merely because agents are fashionable; the scenario's controlling requirement is inspectable task-state graph, not the related framework selection by need concern.

Question 46

A simple three-call assistant is being ported to a heavy agent framework. What should justify the migration?

  1. Perform the critic in a separate context with the draft and evaluation rubric.
  2. Isolate session scratch state from durable, reviewed memory records.
  3. Maintain a compact, explicit summary of durable decisions and retrieve details when needed.
  4. Demonstrate a needed capability such as durable state, branching, or observability.

Explanation

Correct Answer

D:

Framework overhead is worthwhile when its orchestration features solve a real problem, not merely because agents are fashionable.

Incorrect Answers

A:

Isolated critique reduces anchoring to exploratory reasoning and yields a more independent check of the artifact; a remedy for independent evaluator context is not a substitute for framework selection by need.

B:

Temporary observations should expire with the task; promoting only durable facts reduces stale or sensitive data accumulation; even if transient versus durable memory succeeds, framework selection by need still needs separate verification.

C:

Separating long-term task facts from raw transcript prevents context waste while preserving the information needed for new decisions; when the goal is framework selection by need, optimizing compact durable memory misses that operational boundary.

Question 47

A compliance review needs an independent critic that should not inherit all prior draft reasoning. Which structure is useful?

  1. Run the critic in a separate context with the draft and evaluation rubric.
  2. For this workflow, compute aggregates with a tool and pass only validated results back to Claude.
  3. Confirm that the framework exposes enforceable permission or pre-tool hooks.
  4. Partition session scratch state from durable, reviewed memory records.

Explanation

Correct Answer

A:

Isolated critique reduces anchoring to exploratory reasoning and yields a more independent check of the artifact.

Incorrect Answers

B:

Moving mechanical data processing into code avoids token-heavy raw inputs while keeping Claude focused on interpretation; this may help offloading deterministic computation but cannot by itself deliver independent evaluator context.

C:

Framework convenience cannot substitute for mandatory execution controls; approval must intercept the actual write operation; the effect concerns framework approval integration; the success condition is independent evaluator context.

D:

Temporary observations should expire with the task; promoting only durable facts reduces stale or sensitive data accumulation; where independent evaluator context matters, a control for transient versus durable memory leaves that test unmet.

Question 48

An agent handles a user question spanning inventory, shipping, and refunds. How can specialization avoid an oversized prompt?

  1. Separate ownership or serialize writes to shared artifacts.
  2. Route bounded subtasks to domain-specific workers with narrow toolsets.
  3. Request the worker for a bounded evidence summary with references to the originals.
  4. Employ a workflow graph with explicit states and transition rules.

Explanation

Correct Answer

B:

Specialized contexts and tools lower irrelevant information and permission exposure, while a parent assembles the answer.

Incorrect Answers

A:

Parallel reasoning can be safe, but simultaneous edits to common state risk conflicts and lost changes unless ownership is coordinated; the domain-based subagent routing criterion differs from shared-artifact concurrency; the gap persists.

C:

Compact outputs preserve relevant facts and sources while preventing the parent context from being overwhelmed by intermediate material; this may help concise worker return contracts but cannot by itself deliver domain-based subagent routing.

D:

Graph-based orchestration makes conditional edges and retry points inspectable, rather than hiding all sequencing in natural-language planning; where domain-based subagent routing matters, a control for inspectable task-state graph leaves that test unmet.

Question 49

A production tool loop must execute requested client tools and resist duplicate payments after uncertain network failures. Which TWO responsibilities belong to the host application? Choose TWO.

  1. Drop failed tool results from conversation history.
  2. Require an idempotency key for payment execution.
  3. Mark every tool_use message as a completed assistant turn.
  4. Trust the model to avoid reissuing charges after reconnecting.
  5. Dispatch requested tool calls and correlate tool_result blocks by tool_use ID.

Explanation

Correct Answers

B:

A stable key prevents replay of a payment when a timeout leaves the original commit status uncertain. The payment service should store the key with its durable outcome and return that outcome on a repeated attempt.

E:

The model emits proposed calls; the host runs the functions and preserves protocol identifiers for the next turn. For parallel calls, each result must map back to the particular assistant-side request rather than to a guessed call order.

Incorrect Answers

A:

Hiding errors removes the evidence the agent needs to recover safely and could encourage false completion claims. The model needs the failed status to decide whether to retry, ask for help, or stop without claiming success.

C:

A tool request is an intermediate step, not proof that the requested action executed. The host must actually execute the proposed request and report success or failure before the tool round trip is complete.

D:

Conversation content cannot prove which external transaction committed during a timeout. Only transactional state or a deduplication service can resolve a payment whose response was lost.

Question 50

A workflow needs exactly-once-looking external business operations though retries are unavoidable. What combination is most practical?

  1. Use durable state and idempotent tool operations across retries.
  2. Persist the workflow position and completed side effects outside process memory.
  3. Retain a compact, explicit summary of durable decisions and retrieve details when needed.
  4. For this workflow, demonstrate a needed capability such as durable state, branching, or observability.

Explanation

Correct Answer

A:

Recovery requires both knowledge of committed stages and protection from duplicate writes; neither a model summary nor blind retries supplies that guarantee.

Incorrect Answers

B:

Durable checkpoints let the executor resume without blindly rerunning earlier writes after a crash; this targets crash-safe workflow checkpoints, while the system must demonstrate side-effect safe recovery.

C:

Separating long-term task facts from raw transcript prevents context waste while preserving the information needed for new decisions; the effect concerns compact durable memory; the success condition is side-effect safe recovery.

D:

Framework overhead is worthwhile when its orchestration features solve a real problem, not merely because agents are fashionable; even if framework selection by need succeeds, side-effect safe recovery still needs separate verification.

Claude Certified Developer - Foundations (CCDV-F) Practice Tests