Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 19

Question 180

A conversation is saved, but its uploaded attachment expires before the next turn. What should the application retain or handle?

  1. Reopen whichever available attachment has the same filename.
  2. Assume the model permanently memorized the original file.
  3. A valid authorized attachment lifecycle or a clear reattachment path.
  4. Persist the attachment identifier without checking its validity at reuse.

Explanation

Correct Answer

C:

Saved conversation text does not extend the attachment service's retention. Preserve an authorized durable artifact where supported, or explicitly handle expiry and reattachment.

Incorrect Answers

A:

Names are not stable identity or authorization evidence. A replacement must be the intended authorized artifact, not merely similarly named bytes.

B:

Prior processing does not establish durable retrieval of the attachment.

D:

A durable identifier string does not extend the lifetime of its referenced bytes. Reuse needs validity handling or a supported recovery path.

Question 181

Semantic search handles descriptive support questions well but misses an exact product code such as ZX-901Q even though that code is indexed. Which retrieval improvement directly addresses this gap?

  1. Combine semantic retrieval with an exact lexical identifier path.
  2. Raise the semantic similarity threshold for every query.
  3. Ask the answer model to reconstruct the missing product record.
  4. Increase only the answer model's output-token allowance.

Explanation

Correct Answer

A:

Exact codes may carry little semantic meaning. A lexical path can recover literal identifier matches while semantic search continues serving paraphrased natural-language questions.

Incorrect Answers

B:

A stricter threshold can remove more weak semantic matches without adding an exact-code matching mechanism.

C:

A generated guess is not retrieval of the indexed source. It can fabricate details for an identifier the model does not recognize.

D:

The correct record never reaches the answering stage. More generation capacity does not correct retrieval's failure to match an exact indexed code.

Question 182

A document exists in storage but is absent from the search index. Which application boundary should be investigated?

  1. The ingestion and indexing pipeline.
  2. Only whether the source file can be opened manually.
  3. Only the generation prompt after retrieval has already returned no match.
  4. The answer formatter after a successful search.

Explanation

Correct Answer

A:

Storage availability and search discoverability are separate states; index synchronization can fail independently.

Incorrect Answers

B:

File existence proves storage availability, but indexing may still be pending, failed, or scoped differently.

C:

A generation change cannot repair a document omitted from the index or an incomplete ingestion pipeline.

D:

Formatting occurs after retrieval and cannot explain why an existing document never entered the searchable index.

Question 183

Plugin A must finish initializing before B, B before C, and C before A. All three are required. How should the application handle this dependency configuration?

  1. Start alphabetically and assume the dependencies will resolve later.
  2. Detect the cycle and require the dependencies to be redesigned.
  3. Mark one required plugin initialized without running it.
  4. Retry the same sequential initialization indefinitely.

Explanation

Correct Answer

B:

There is no valid initialization order satisfying all three prerequisites. A cycle check exposes a configuration defect that cannot be repaired by choosing a different first plugin.

Incorrect Answers

A:

Alphabetical order does not satisfy the circular prerequisites. At least one plugin would start before a required dependency.

C:

A fabricated state can allow dependents to start against missing functionality, hiding rather than resolving the invalid graph.

D:

The graph remains cyclic on every attempt. Waiting cannot create a valid ordering unless the dependency declarations change.

Question 184

Two installed integrations register the identical dispatch name search. One registration silently replaces the other. What should the registry do?

  1. Keep whichever implementation was installed most recently without notice.
  2. Choose an implementation randomly for each call.
  3. Require unambiguous identifiers or reject the collision.
  4. Send each search call to both services automatically.

Explanation

Correct Answer

C:

Dispatch identity must resolve to one intended implementation. Namespacing or an explicit collision error prevents installation order from silently choosing which service receives calls.

Incorrect Answers

A:

That makes existing calls change destination after an unrelated installation and leaves the caller unaware of the replaced contract.

B:

Random dispatch cannot preserve the named tool's promised behavior or allow a caller to target a specific service reliably.

D:

Fan-out changes the operation's cost, data exposure, and result semantics. It is not a safe implicit resolution of a registry collision.

Question 185

A user asks for concise answers in one task. What must be explicit before treating that as a permanent preference?

  1. Every temporary instruction must become permanent memory.
  2. Whether the instruction is task-local or intended for durable personalization.
  3. Store it under a shared global user record.
  4. Never allow users to update stored preferences.

Explanation

Correct Answer

B:

Scope determines whether later unrelated tasks should inherit the preference; persistence should not be inferred indiscriminately.

Incorrect Answers

A:

That can propagate context-specific choices beyond their intended scope.

C:

Global storage mixes different users’ preferences and ownership.

D:

A durable preference still needs a controlled correction mechanism.

Question 186

An application routes answers using a model’s self-reported confidence. What must be checked before trusting the threshold?

  1. That every self-reported score is numerically between zero and one.
  2. That repeated calls produce the same self-reported confidence.
  3. Calibration against labeled outcomes on representative data.
  4. That the threshold matches a value used for a different task.

Explanation

Correct Answer

C:

Compare confidence values with observed correctness on representative labeled data. This establishes whether the routing threshold predicts the intended outcome; numerical precision alone does not.

Incorrect Answers

A:

A valid numeric range establishes format, not whether a score predicts correctness at the chosen threshold.

B:

Repeatability is not calibration: a consistently confident model can be consistently wrong. The threshold needs labeled outcome evidence.

D:

Calibration depends on the workload and outcomes; an unrelated threshold does not establish this router's reliability.

Question 187

Two providers represent errors and usage differently. What should a shared application adapter preserve?

  1. A common internal contract without losing provider-specific meaning needed for recovery.
  2. Map all provider errors into one retryable failure category.
  3. Discard all usage fields because names differ.
  4. Assume identical field names imply identical semantics.

Explanation

Correct Answer

A:

Normalization should simplify callers while retaining distinctions that affect retries, diagnostics, and accounting.

Incorrect Answers

B:

Collapsing invalid input, authorization, and transient failures removes distinctions that determine whether retrying is appropriate.

C:

Different schemas require mapping, not abandonment of resource accounting.

D:

Names alone do not prove equivalent meaning across interfaces.

Question 188

A model returns text containing HTML-like markup. The product intends to display plain text. What rendering choice fits?

  1. Remove script tags and insert all remaining markup as trusted HTML.
  2. Treat the response as trusted because it came through an authenticated API.
  3. Escape or safely render it as text rather than executing markup.
  4. Insert it directly as trusted innerHTML.

Explanation

Correct Answer

C:

Output is data; the display layer must preserve the intended content boundary instead of introducing executable browser behavior.

Incorrect Answers

A:

A tag blacklist is not the intended plain-text rendering contract and may miss other active markup. Text-safe rendering avoids interpreting the content.

B:

Provider authentication does not turn generated text into trusted browser markup. The output still needs the display boundary required by the product.

D:

That can interpret untrusted generated content as active markup.

Question 189

A user requests “update Alex’s record,” and three authorized records match. What should the application do?

  1. Update all matching records.
  2. Obtain enough distinguishing information before proposing the update.
  3. Select the oldest record automatically.
  4. Use the first database result without checking how results are ordered.

Explanation

Correct Answer

B:

Multiple valid matches make the target ambiguous; a mutation should not depend on an arbitrary identity guess.

Incorrect Answers

A:

The singular request does not authorize a bulk modification.

C:

Record age does not establish which Alex the user intended.

D:

Result ordering is not evidence of which Alex the user intended, and can change independently of identity.

Question 190

A UI labels a model’s proposed action “Completed” before its tool finishes. What should change?

  1. Separate proposed, executing, succeeded, and failed states.
  2. Mark success when the dispatcher accepts the request into its queue.
  3. Keep one Completed state and place failures only in server logs.
  4. Mark completion when the model emits a syntactically valid tool request.

Explanation

Correct Answer

A:

Drive completion from the executor's verified outcome. Separate proposal, execution, success, and failure so the interface does not present a planned change as an accomplished one.

Incorrect Answers

B:

Queue admission establishes that execution was accepted, not that the external action completed successfully. Completion must follow the outcome.

C:

A single success label hides failure from the user. The interface needs states that communicate actual execution outcomes.

D:

A valid request is still a proposal to the executor; it does not establish that the external operation succeeded.

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