Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 44

Question 430

A design review identifies a missing capability in an inference routing project: it must smooth sudden ingestion spikes before workers call a capacity-limited model API. Which change meets that goal?

  1. Choose WebSockets for bidirectional realtime exchange
  2. Send the correct API authentication header
  3. Use a queue for burst absorption
  4. Send an API version header where required

Explanation

Correct Answer

C:

Queueing decouples producers from consumer throughput and enables controlled retries; this behavior directly supports the need to smooth sudden ingestion spikes before workers call a capacity-limited model API. Increasing frontend timeouts does not eliminate backend capacity limits.

Incorrect Answers

A:

This approach serves the separate need to implement an interactive collaboration channel requiring messages in both directions because persistent two-way channels fit interactive event exchange where the client also sends messages during a session; it does not establish the outcome to smooth sudden ingestion spikes before workers call a capacity-limited model API, and standard response streaming alone is not a general bidirectional messaging protocol.

B:

The API validates the provided credential before accepting a request; despite being useful when a service must avoid unauthenticated model requests from a backend worker, it is not a solution to smooth sudden ingestion spikes before workers call a capacity-limited model API; embedding a secret in a user prompt does not authenticate an HTTP call.

D:

Explicit protocol versioning makes the caller's expected API semantics unambiguous; this applies to the different goal to keep integration behavior aligned with a supported request contract, not the requirement to smooth sudden ingestion spikes before workers call a capacity-limited model API; relying on an arbitrary undocumented default is fragile.

Question 431

During a backend integration rollout, two different failure paths need remediation. The first component must keep external tool connectors compatible after client upgrades; separately, the second must fail fast if a model ID or tool endpoint is absent in production. Which changes should be adopted?

  1. First component: Keep environment-specific settings external. Second component: Roll out configuration through reviewed changes.
  2. First component: Track MCP server versions and contracts. Second component: Validate configuration at startup.
  3. First component: Roll out configuration through reviewed changes. Second component: Validate configuration at startup.
  4. First component: Track MCP server versions and contracts. Second component: Keep environment-specific settings external.

Explanation

Correct Answer

B:

The first proposal fits its assigned role because declared versions and integration tests expose resource or schema changes before deployment, while the second proposal fits its assigned role because startup checks surface missing required values before user traffic reaches a broken path; their respective acceptance criteria are to keep external tool connectors compatible after client upgrades and to fail fast if a model ID or tool endpoint is absent in production.

Incorrect Answers

A:

The first proposal serves the different purpose to deploy the same application build to staging and production with distinct endpoints instead of the assigned first role because configuration injection separates environment differences from immutable application artifacts, while the second proposal serves the different purpose to prevent an untested tool-allowlist edit from reaching all developers immediately instead of the assigned second role because change review and versioning provide accountability and a tested rollback path; their respective acceptance criteria are to keep external tool connectors compatible after client upgrades and to fail fast if a model ID or tool endpoint is absent in production.

C:

The first proposal serves the different purpose to prevent an untested tool-allowlist edit from reaching all developers immediately instead of the assigned first role because change review and versioning provide accountability and a tested rollback path, while the second proposal fits its assigned role because startup checks surface missing required values before user traffic reaches a broken path; their respective acceptance criteria are to keep external tool connectors compatible after client upgrades and to fail fast if a model ID or tool endpoint is absent in production.

D:

The first proposal fits its assigned role because declared versions and integration tests expose resource or schema changes before deployment, while the second proposal serves the different purpose to deploy the same application build to staging and production with distinct endpoints instead of the assigned second role because configuration injection separates environment differences from immutable application artifacts; their respective acceptance criteria are to keep external tool connectors compatible after client upgrades and to fail fast if a model ID or tool endpoint is absent in production.

Question 432

An engineer extending an MCP interoperability workflow has been asked to identify whether slow agent replies come from a payment gateway rather than the model. Which choice addresses the request most precisely?

  1. Add an approval boundary before sensitive writes
  2. Restrict tool network credentials
  3. Write task-specific tool descriptions
  4. Capture tool latency and failure metrics

Explanation

Correct Answer

D:

Per-tool tracing separates external execution time from inference and highlights unhealthy dependencies; this behavior directly supports the need to identify whether slow agent replies come from a payment gateway rather than the model. Aggregate response time alone cannot identify the responsible connector.

Incorrect Answers

A:

Authorization by a user or operator can be enforced before the executor commits changes; despite being useful when a service must prevent a helpful assistant from issuing refunds automatically, it is not a solution to identify whether slow agent replies come from a payment gateway rather than the model; a note in the model's prompt is not a transaction approval.

B:

For the alternate objective to stop a directory lookup function from also gaining unrestricted database write access, minimal capabilities reduce risk when generated arguments or retrieved text are hostile; the feature here must identify whether slow agent replies come from a payment gateway rather than the model, and better prompt wording does not remove excess infrastructure permissions.

C:

Clear capabilities and side effects guide tool selection without granting new authority; this applies to the different goal to let Claude distinguish a read-only invoice lookup from an invoice update, not the requirement to identify whether slow agent replies come from a payment gateway rather than the model; vague overlapping descriptions increase erroneous calls.

Question 433

A design review identifies a missing capability in a prompt engineering project: it must prevent a parser crash when the model returns no ordinary text. Which change meets that goal?

  1. Separate a partial streaming fragment from a finished object
  2. Escape generated content in the UI
  3. Validate required schema fields
  4. Handle refusal and empty content explicitly

Explanation

Correct Answer

D:

Response handling must branch on stop reason and content block types before decoding data; this behavior directly supports the need to prevent a parser crash when the model returns no ordinary text. Assuming every call yields a usable business record creates hidden error paths.

Incorrect Answers

A:

This approach serves the separate need to prevent the frontend from parsing an incomplete JSON chunk as a complete response because stream consumers must buffer and recognize completion before strict object validation; it does not establish the outcome to prevent a parser crash when the model returns no ordinary text, and an arbitrary content delta may end halfway through a field.

B:

Output encoding or sanitization protects display contexts from untrusted markup; despite being useful when a service must show model-produced HTML snippets without allowing browser script injection, it is not a solution to prevent a parser crash when the model returns no ordinary text; a correct JSON wrapper does not make embedded HTML safe.

C:

For the alternate objective to reject a response that omits a mandated account identifier, schema checks enforce field presence, types, and enumerations before the result enters a database; the feature here must prevent a parser crash when the model returns no ordinary text, and syntactically valid JSON may still be contract-invalid.

Question 434

Engineers are splitting a document processing service into two independently tested functions. Function A must prevent abandoned connectors from retaining access after an application is transferred, while function B must test whether document classification quality is viable on representative files. Which plan is sound?

  1. First component: Perform a threat review before go-live. Second component: Define a staged release with rollback.
  2. First component: Assign an owner to every integration. Second component: Perform a threat review before go-live.
  3. First component: Assign an owner to every integration. Second component: Use a proof of concept before production commitment.
  4. First component: Define a staged release with rollback. Second component: Use a proof of concept before production commitment.

Explanation

Correct Answer

C:

The first proposal fits its assigned role because ownership enables rotation, monitoring, change approval, and retirement decisions, while the second proposal fits its assigned role because a small pilot reduces feasibility risk before funding a complete integration; their respective acceptance criteria are to prevent abandoned connectors from retaining access after an application is transferred and to test whether document classification quality is viable on representative files.

Incorrect Answers

A:

The first proposal serves the different purpose to uncover prompt-injection paths through retrieved webpages and connected tools instead of the assigned first role because reviewing trust boundaries and permissions before deployment reveals attack routes unit tests may miss, while the second proposal serves the different purpose to deploy a prompt change without exposing every customer to an untested regression instead of the assigned second role because gradual rollout and versioned rollback allow metrics to be compared while limiting impact; their respective acceptance criteria are to prevent abandoned connectors from retaining access after an application is transferred and to test whether document classification quality is viable on representative files.

B:

The first proposal fits its assigned role because ownership enables rotation, monitoring, change approval, and retirement decisions, while the second proposal serves the different purpose to uncover prompt-injection paths through retrieved webpages and connected tools instead of the assigned second role because reviewing trust boundaries and permissions before deployment reveals attack routes unit tests may miss; their respective acceptance criteria are to prevent abandoned connectors from retaining access after an application is transferred and to test whether document classification quality is viable on representative files.

D:

The first proposal serves the different purpose to deploy a prompt change without exposing every customer to an untested regression instead of the assigned first role because gradual rollout and versioned rollback allow metrics to be compared while limiting impact, while the second proposal fits its assigned role because a small pilot reduces feasibility risk before funding a complete integration; their respective acceptance criteria are to prevent abandoned connectors from retaining access after an application is transferred and to test whether document classification quality is viable on representative files.

Question 435

An output validation group is preparing a production change intended to prevent a customer-facing assistant from making final legal determinations without review. Which technique matches that intent?

  1. Use explicit refusal and escalation instructions
  2. Use few-shot boundary cases
  3. State one measurable outcome per instruction
  4. Constrain response audience and purpose
  5. Specify output format in the prompt

Explanation

Correct Answer

A:

Clear boundaries describe when to abstain or route to a qualified person; this behavior directly supports the need to prevent a customer-facing assistant from making final legal determinations without review. Text instructions require supporting application enforcement for high-risk decisions.

Incorrect Answers

B:

This approach serves the separate need to distinguish near-duplicate support intents that a broad rule routinely confuses because examples illustrating each side of a decision boundary communicate nuanced classification criteria; it does not establish the outcome to prevent a customer-facing assistant from making final legal determinations without review, and many unrelated demonstrations can make the rule harder to infer.

C:

Clear task goals and acceptance criteria reduce competing interpretations; despite being useful when a service must stop a summarizer from producing unrelated analysis instead of required action items, it is not a solution to prevent a customer-facing assistant from making final legal determinations without review; a vague request for a good summary provides no testable target.

D:

For the alternate objective to generate a technical incident summary suitable for an on-call engineer instead of an executive, audience constraints determine useful terminology, specificity, and level of detail; the feature here must prevent a customer-facing assistant from making final legal determinations without review, and style guidance alone cannot make inaccurate facts true.

E:

Explicit field descriptions guide generation toward the consumer's expected structure; this applies to the different goal to obtain named fields in a consistent response layout for subsequent processing, not the requirement to prevent a customer-facing assistant from making final legal determinations without review; formatting guidance still requires machine validation at the boundary.

Question 436

A technical lead requests a solution that can send extracted contract fields into a database that rejects unexpected shapes within a backend integration project. Which design should be chosen?

  1. Isolate tools behind a service layer
  2. Add explicit provenance to retrieved evidence
  3. Use bounded retrieval results
  4. Build a typed output envelope

Explanation

Correct Answer

D:

An explicit schema with required fields, types, and allowed values makes integration errors detectable; this behavior directly supports the need to send extracted contract fields into a database that rejects unexpected shapes. Free-form natural language is difficult to validate automatically.

Incorrect Answers

A:

This approach serves the separate need to avoid exposing raw production database credentials directly to an agent because a mediation service validates identity, arguments, authorization, and audit logs for each requested action; it does not establish the outcome to send extracted contract fields into a database that rejects unexpected shapes, and a natural-language tool description is not an access-control boundary.

B:

Source identifiers, timestamps, and passage boundaries make citations and later audits possible; despite being useful when a service must let reviewers trace each summary claim to its authoritative document version, it is not a solution to send extracted contract fields into a database that rejects unexpected shapes; a fluent answer without lineage cannot be independently verified.

C:

For the alternate objective to avoid burying the relevant contract clause beneath hundreds of low-value passages, ranking and a context budget prioritize high-quality evidence while controlling input tokens; the feature here must send extracted contract fields into a database that rejects unexpected shapes, and sending every possible hit can cause context bloat and distract attention.

Question 437

A technical lead requests a solution that can avoid confusing allowable prompt size with maximum generated response length within a token economics project. Which design should be chosen?

  1. Understand context-window competition
  2. Use varied sampling only for ideation
  3. Budget tokens rather than characters
  4. Distinguish input from output token limits

Explanation

Correct Answer

D:

The context budget constrains total material considered while output ceilings bound what a single generation can emit; this behavior directly supports the need to avoid confusing allowable prompt size with maximum generated response length. Raising an output cap does not make an oversized input fit.

Incorrect Answers

A:

When an application needs to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity, all supplied content competes for a finite effective context budget and may affect attention; the current capability must instead avoid confusing allowable prompt size with maximum generated response length, while a larger prompt is not inherently more accurate.

B:

Greater sampling diversity can produce a wider range of suggestions for exploratory work; the supported outcome is to brainstorm diverse naming candidates without demanding identical wording on every run, whereas this scenario requires the ability to avoid confusing allowable prompt size with maximum generated response length; it is not appropriate as the sole correctness control for structured production data.

C:

This approach serves the separate need to estimate whether a multilingual request will fit a model context window because tokenization splits text into model-specific units, so character counts are not a reliable context proxy; it does not establish the outcome to avoid confusing allowable prompt size with maximum generated response length, and equal-length strings can have different token usage.

Question 438

Engineers are splitting a backend integration service into two independently tested functions. Function A must require manual signoff when an AI-produced recommendation exceeds a financial threshold, while function B must prevent a Claude assistant from inventing account balances when it lacks current ledger data. Which plan is sound?

  1. First component: Specify human-review triggers. Second component: Identify authoritative source systems.
  2. First component: Set an explicit data residency constraint. Second component: Identify authoritative source systems.
  3. First component: Classify data sensitivity up front. Second component: Set an explicit data residency constraint.
  4. First component: Specify human-review triggers. Second component: Classify data sensitivity up front.

Explanation

Correct Answer

A:

The first proposal fits its assigned role because decision thresholds turn governance expectations into testable escalation rules, while the second proposal fits its assigned role because source-of-truth mapping identifies which verified systems must supply facts and what freshness guarantees apply; their respective acceptance criteria are to require manual signoff when an AI-produced recommendation exceeds a financial threshold and to prevent a Claude assistant from inventing account balances when it lacks current ledger data.

Incorrect Answers

B:

The first proposal serves the different purpose to ensure protected records are processed only in approved jurisdictions instead of the assigned first role because documented residency requirements inform cloud placement, logging, storage, and vendor selection, while the second proposal fits its assigned role because source-of-truth mapping identifies which verified systems must supply facts and what freshness guarantees apply; their respective acceptance criteria are to require manual signoff when an AI-produced recommendation exceeds a financial threshold and to prevent a Claude assistant from inventing account balances when it lacks current ledger data.

C:

The first proposal serves the different purpose to decide which customer fields must be redacted before model requests instead of the assigned first role because data classification guides minimization, access controls, retention, and acceptable transfers, while the second proposal serves the different purpose to ensure protected records are processed only in approved jurisdictions instead of the assigned second role because documented residency requirements inform cloud placement, logging, storage, and vendor selection; their respective acceptance criteria are to require manual signoff when an AI-produced recommendation exceeds a financial threshold and to prevent a Claude assistant from inventing account balances when it lacks current ledger data.

D:

The first proposal fits its assigned role because decision thresholds turn governance expectations into testable escalation rules, while the second proposal serves the different purpose to decide which customer fields must be redacted before model requests instead of the assigned second role because data classification guides minimization, access controls, retention, and acceptable transfers; their respective acceptance criteria are to require manual signoff when an AI-produced recommendation exceeds a financial threshold and to prevent a Claude assistant from inventing account balances when it lacks current ledger data.

Question 439

A technical specification for a language-model integration application assigns two jobs to separate modules: module A must compare two models where the cheaper call fails validation more often, and module B must classify many short messages where low latency and unit cost dominate. Which two-part solution is appropriate?

  1. First component: Balance quality against price per successful task. Second component: Measure end-to-end latency, not just token speed.
  2. First component: Measure end-to-end latency, not just token speed. Second component: Prefer a Sonnet-class model for balanced workloads.
  3. First component: Prefer a Sonnet-class model for balanced workloads. Second component: Prefer a Haiku-class model for lightweight routing.
  4. First component: Balance quality against price per successful task. Second component: Prefer a Haiku-class model for lightweight routing.

Explanation

Correct Answer

D:

The first proposal fits its assigned role because retries, human review, and downstream error cost can outweigh nominal token-price savings, while the second proposal fits its assigned role because the lighter tier suits inexpensive, high-throughput tasks with modest reasoning demand; their respective acceptance criteria are to compare two models where the cheaper call fails validation more often and to classify many short messages where low latency and unit cost dominate.

Incorrect Answers

A:

The first proposal fits its assigned role because retries, human review, and downstream error cost can outweigh nominal token-price savings, while the second proposal serves the different purpose to select the right model for an application with tool and network overhead instead of the assigned second role because total response time includes queuing, input processing, generation, and downstream integration; their respective acceptance criteria are to compare two models where the cheaper call fails validation more often and to classify many short messages where low latency and unit cost dominate.

B:

The first proposal serves the different purpose to select the right model for an application with tool and network overhead instead of the assigned first role because total response time includes queuing, input processing, generation, and downstream integration, while the second proposal serves the different purpose to serve a mixed coding and support assistant at moderate latency and cost instead of the assigned second role because the balanced tier often offers a practical quality-throughput compromise, subject to task-level testing; their respective acceptance criteria are to compare two models where the cheaper call fails validation more often and to classify many short messages where low latency and unit cost dominate.

C:

The first proposal serves the different purpose to serve a mixed coding and support assistant at moderate latency and cost instead of the assigned first role because the balanced tier often offers a practical quality-throughput compromise, subject to task-level testing, while the second proposal fits its assigned role because the lighter tier suits inexpensive, high-throughput tasks with modest reasoning demand; their respective acceptance criteria are to compare two models where the cheaper call fails validation more often and to classify many short messages where low latency and unit cost dominate.

Question 440

During a language-model integration rollout, two different failure paths need remediation. The first component must estimate whether a multilingual request will fit a model context window; separately, the second must explain why a plausible answer can still contain an unsupported factual claim. Which changes should be adopted?

  1. First component: Budget tokens rather than characters. Second component: Recognize next-token generation.
  2. First component: Use varied sampling only for ideation. Second component: Recognize next-token generation.
  3. First component: Understand context-window competition. Second component: Use varied sampling only for ideation.
  4. First component: Budget tokens rather than characters. Second component: Understand context-window competition.

Explanation

Correct Answer

A:

The first proposal fits its assigned role because tokenization splits text into model-specific units, so character counts are not a reliable context proxy, while the second proposal fits its assigned role because the model predicts likely continuations conditioned on context rather than querying an authoritative truth database by default; their respective acceptance criteria are to estimate whether a multilingual request will fit a model context window and to explain why a plausible answer can still contain an unsupported factual claim.

Incorrect Answers

B:

The first proposal serves the different purpose to brainstorm diverse naming candidates without demanding identical wording on every run instead of the assigned first role because greater sampling diversity can produce a wider range of suggestions for exploratory work, while the second proposal fits its assigned role because the model predicts likely continuations conditioned on context rather than querying an authoritative truth database by default; their respective acceptance criteria are to estimate whether a multilingual request will fit a model context window and to explain why a plausible answer can still contain an unsupported factual claim.

C:

The first proposal serves the different purpose to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity instead of the assigned first role because all supplied content competes for a finite effective context budget and may affect attention, while the second proposal serves the different purpose to brainstorm diverse naming candidates without demanding identical wording on every run instead of the assigned second role because greater sampling diversity can produce a wider range of suggestions for exploratory work; their respective acceptance criteria are to estimate whether a multilingual request will fit a model context window and to explain why a plausible answer can still contain an unsupported factual claim.

D:

The first proposal fits its assigned role because tokenization splits text into model-specific units, so character counts are not a reliable context proxy, while the second proposal serves the different purpose to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity instead of the assigned second role because all supplied content competes for a finite effective context budget and may affect attention; their respective acceptance criteria are to estimate whether a multilingual request will fit a model context window and to explain why a plausible answer can still contain an unsupported factual claim.

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