Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 48

Question 470

A desktop application can call an MCP server locally while another runtime uses a remote server. Which approach targets the deciding condition?

  1. Use stdio for a local subprocess integration
  2. Use Streamable HTTP for remote MCP access
  3. Use a client-side transport adapter
  4. Expose a custom API adapter outside MCP when local-only

Explanation

Correct Answer

C:

Transport abstraction lets the same protocol-level concepts operate over supported connection methods, although a transport switch does not erase server deployment or security requirements.

Incorrect Answers

A:

Standard input and output suit tightly coupled local client-server interactions; by comparison, the intended solution relies on this: transport abstraction lets the same protocol-level concepts operate over supported connection methods; the server's standard output must remain protocol-clean instead of mixing human debug prints.

B:

A network transport supports remote requests under explicit authentication and deployment controls; the needed capability is different—transport abstraction lets the same protocol-level concepts operate over supported connection methods—and opening an HTTP endpoint without access restrictions can expose powerful tools publicly.

D:

A direct adapter may be simpler than running and maintaining a separate protocol server; however, the critical mechanism here is that transport abstraction lets the same protocol-level concepts operate over supported connection methods, and choosing MCP solely for naming consistency introduces unnecessary lifecycle work.

Question 471

A shared search system holds confidential documents from multiple teams and users have distinct permissions. Which capability is necessary for this outcome?

  1. Retrieve only topically relevant passages
  2. Refresh an index when content changes
  3. Validate the caller's identity before exposing tools
  4. Apply authorization before retrieval

Explanation

Correct Answer

D:

The retrieval layer filters sources by verified caller access before any text reaches Claude; in production, filtering only the final answer cannot undo earlier disclosure to the model.

Incorrect Answers

A:

Search or retrieval supplies a bounded evidence set instead of the entire collection; this addresses a separate concern, not the fact that the retrieval layer filters sources by verified caller access before any text reaches Claude; sending all documents raises cost, noise, and exposure without improving source precision.

B:

A freshness pipeline updates search representations and source versions when documents change; here, the controlling capability is that the retrieval layer filters sources by verified caller access before any text reaches Claude, but more prompt instructions cannot correct an out-of-date underlying index.

C:

Authentication establishes which principal is making the request before access decisions; the missing capability is more specific: the retrieval layer filters sources by verified caller access before any text reaches Claude; a conversational claim of ownership is not reliable proof of identity.

Question 472

An infinite agent loop or repeated giant prompt suddenly increases usage far above baseline. What is the right implementation response?

  1. Alert on unusual token spikes
  2. Compute cost from input and output token usage
  3. Tie cost optimization to measured quality
  4. Forecast p95 cost rather than just average

Explanation

Correct Answer

A:

Cost monitoring can detect and halt anomalous demand before it causes significant overspend; the important constraint is that monthly billing review alone detects runaway costs too late.

Incorrect Answers

B:

Per-model input and output token rates combine with measured usage for accurate costing; the requirement instead turns on this behavior: cost monitoring can detect and halt anomalous demand before it causes significant overspend; number of messages alone is not an adequate unit of resource consumption.

C:

Total cost analysis includes downstream review and failure impact, not just provider charges; this cannot substitute for the behavior that cost monitoring can detect and halt anomalous demand before it causes significant overspend, because a cheaper inference call is not a net saving if error handling multiplies workload.

D:

Distribution-aware forecasting accounts for long-tail tasks and tool iterations; in contrast, cost monitoring can detect and halt anomalous demand before it causes significant overspend, while median request cost can hide budget risk from heavy cases.

Question 473

An application receives simple FAQs and rare difficult investigations under one interface. What is the best fit for the described situation?

  1. Use a staged escalation strategy
  2. Route dynamically among evaluated tiers
  3. Prefer Opus-class models for demanding reasoning
  4. Prefer Haiku-class models for validated simple tasks

Explanation

Correct Answer

B:

A routing policy can send easy cases to economical models and escalate complex work; misclassification by the router can lower quality on unexpectedly difficult tasks.

Incorrect Answers

A:

An escalation policy retries only the genuinely hard or failed cases on a higher-capability tier; by comparison, the intended solution relies on this: a routing policy can send easy cases to economical models and escalate complex work; escalation thresholds must be evaluated to avoid excessive retries.

C:

The top capability tier is a candidate when advanced reasoning quality dominates the tradeoff, whereas the requirement calls for another behavior: a routing policy can send easy cases to economical models and escalate complex work; higher cost and time should be justified through measured improvement.

D:

A smaller model family often minimizes latency and unit cost for routine operations; however, the critical mechanism here is that a routing policy can send easy cases to economical models and escalate complex work, and actual suitability should be verified with representative task evaluations.

Question 474

A high-volume text categorizer reaches its quality target with a fast lower-cost Claude variant. Which capability is necessary for this outcome?

  1. Use a capable tier for complex synthesis
  2. Prefer Haiku-class models for validated simple tasks
  3. Prefer Sonnet-class models for balanced workloads
  4. Route dynamically among evaluated tiers

Explanation

Correct Answer

B:

A smaller model family often minimizes latency and unit cost for routine operations—with this limitation: actual suitability should be verified with representative task evaluations.

Incorrect Answers

A:

A stronger model may justify its cost when deeper synthesis materially improves decisions; however, the critical mechanism here is that a smaller model family often minimizes latency and unit cost for routine operations, and its result still requires evidence and domain-specific review.

C:

A balanced model tier can provide strong capability at lower overhead than the highest tier; this cannot substitute for the behavior that a smaller model family often minimizes latency and unit cost for routine operations, because higher complexity may still justify a more powerful model.

D:

A routing policy can send easy cases to economical models and escalate complex work; in contrast, a smaller model family often minimizes latency and unit cost for routine operations, while misclassification by the router can lower quality on unexpectedly difficult tasks.

Question 475

A document processing task can take minutes and should survive the original HTTP request closing. What should the team do at this decision point?

  1. Use a work queue for long offline jobs
  2. Distinguish model-call retry from tool-action replay
  3. Bound concurrency with a semaphore
  4. Cancel timed-out tasks cleanly

Explanation

Correct Answer

A:

A persistent queue decouples submission from workers and supports retry state—with this limitation: keeping a web request open indefinitely is fragile under client disconnects.

Incorrect Answers

B:

Separating generation retries from durable side-effect records avoids duplicate tool execution; this cannot substitute for the behavior that a persistent queue decouples submission from workers and supports retry state, because model text has no transactional guarantee that would make action replays harmless.

C:

A concurrency limit caps simultaneous API work and prevents uncontrolled bursts; in contrast, a persistent queue decouples submission from workers and supports retry state, while launching every coroutine at once can trigger rate limiting and memory pressure.

D:

Cancellation and timeout handling release resources and prevent orphaned work where supported; this addresses a separate concern, not the fact that a persistent queue decouples submission from workers and supports retry state; merely hiding the client spinner does not stop backend consumption.

Question 476

Finance wants to compare agent runs that produce equivalent summaries at different costs. Which design would meet this acceptance criterion?

  1. Record tool-call arguments and outcomes safely
  2. Tag each step with a correlation identifier
  3. Dispatch each returned tool_use by its name
  4. Track token and tool budgets per task

Explanation

Correct Answer

D:

Metering links request usage and tool counts to the logical job or trace identifier; the important constraint is that a single final answer's length is not a complete measure of work performed.

Incorrect Answers

A:

Structured traces capture call names, sanitized inputs, responses, timing, and error classification; by comparison, the intended solution relies on this: metering links request usage and tool counts to the logical job or trace identifier; logging must redact secrets and sensitive payload fields before persistence.

B:

Propagating a stable trace ID across calls permits cause-and-effect analysis across services; the needed capability is different—metering links request usage and tool counts to the logical job or trace identifier—and relying on timestamps alone makes concurrent sessions difficult to reconstruct.

C:

The harness validates the tool identifier and arguments before invoking the corresponding local function, whereas the requirement calls for another behavior: metering links request usage and tool counts to the logical job or trace identifier; the content block is a request to run a tool, not the executed inventory result.

Question 477

A policy assistant holds a large handbook but usually needs just a few matching sections for each answer. How should this requirement be met?

  1. Remove stale tool outputs
  2. Retrieve relevant evidence per user question
  3. Isolate unrelated subtasks in separate contexts
  4. Track source timestamps or versions

Explanation

Correct Answer

B:

Retrieval selects bounded passages pertinent to the current question and supplies their source context; the implementation must still recognize that selecting on superficial keywords alone may miss nuanced exceptions.

Incorrect Answers

A:

Pruning outdated results frees context for more relevant evidence and current questions; this addresses a separate concern, not the fact that retrieval selects bounded passages pertinent to the current question and supplies their source context; removing the sole authoritative citation would weaken the final answer.

C:

Distinct subagent contexts prevent irrelevant data from filling the coordinator's main conversation; the missing capability is more specific: retrieval selects bounded passages pertinent to the current question and supplies their source context; true confidentiality still depends on tool access and data-sharing controls.

D:

Version metadata allows the agent to prefer current evidence and flag historical material; by comparison, the intended solution relies on this: retrieval selects bounded passages pertinent to the current question and supplies their source context; timestamp alone does not establish which source is authoritative.

Question 478

A developer asks about the current API contract and old conversational memories may be stale. How should the implementation handle this need?

  1. Use retrieval from a source repository
  2. Use a compact task summary for handoff
  3. Use a fallback to a human decision-maker
  4. Use keyed long-term memory for approved facts

Explanation

Correct Answer

A:

Fetching current documentation from a maintained source avoids depending on recollected assertions, subject to this requirement: retrieved snippets still need source and version checks before action.

Incorrect Answers

B:

A narrow summary communicates findings, evidence, and unresolved questions within a smaller token budget; here, the controlling capability is that fetching current documentation from a maintained source avoids depending on recollected assertions, but it must not silently claim unverified findings are authoritative.

C:

Escalation transfers the ambiguous decision with relevant evidence and uncertainties; the missing capability is more specific: fetching current documentation from a maintained source avoids depending on recollected assertions; generating a confident answer is not an acceptable replacement for the required human approval.

D:

Stored records indexed by authorized account ID can be retrieved for later tasks; by comparison, the intended solution relies on this: fetching current documentation from a maintained source avoids depending on recollected assertions; access controls and deletion policy must govern persistence.

Question 479

A backend sends independent prompt messages and expects one finite answer per request. What is the best fit for the described situation?

  1. Use HTTP request-response for discrete inference
  2. Use a typed abstraction for business-specific inference
  3. Use WebSockets for application-managed duplex traffic
  4. Use a queue for disconnected long-running workloads

Explanation

Correct Answer

A:

Standard HTTPS requests fit isolated generation operations without long-lived bidirectional connections, although opening a persistent socket adds state and lifecycle complexity where it is unnecessary.

Incorrect Answers

B:

A domain wrapper hides model-request details behind validated inputs and outputs; however, the critical mechanism here is that standard HTTPS requests fit isolated generation operations without long-lived bidirectional connections, and abstractions should expose meaningful errors instead of swallowing model uncertainty.

C:

A WebSocket supports messages in both directions on one persistent connection; the requirement instead turns on this behavior: standard HTTPS requests fit isolated generation operations without long-lived bidirectional connections; its availability does not imply the standard Claude Messages endpoint is a WebSocket API.

D:

Durable queued work survives client disconnection and exposes later status; this cannot substitute for the behavior that standard HTTPS requests fit isolated generation operations without long-lived bidirectional connections, because keeping an open WebSocket alone does not make background jobs persistent.

Question 480

A regulated archival platform must require two separate authorized custodians to approve permanent bulk erasure after the requesting customer has confirmed intent. Which choice should be used here?

  1. Use transaction controls for critical writes
  2. Audit privileged changes with actor attribution
  3. Record immutable provenance for sensitive decisions
  4. Require two independent approvers for regulated bulk deletion

Explanation

Correct Answer

D:

A dual-control workflow with separate identities prevents one actor or agent from approving the entire irreversible removal; a second confirmation button by the same person does not meet the two-person policy.

Incorrect Answers

A:

Atomic transactions or compensating workflows preserve business consistency across approved changes; here, the controlling capability is that a dual-control workflow with separate identities prevents one actor or agent from approving the entire irreversible removal, but model reasoning does not guarantee database rollback.

B:

An audit trail connects the principal, approved operation, target, and result; the missing capability is more specific: a dual-control workflow with separate identities prevents one actor or agent from approving the entire irreversible removal; a text summary by the model is not sufficient evidence of a committed security change.

C:

Audit metadata preserves input references, applicable policy, actor, and versioned output; by comparison, the intended solution relies on this: a dual-control workflow with separate identities prevents one actor or agent from approving the entire irreversible removal; a final answer without source history is not adequate for reconstructing the decision.

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