Question 110
Old and new application versions must run simultaneously during deployment. A shared structured field is being renamed. Which migration preserves compatibility during the overlap?
- Switch producers immediately and update consumers only when they fail.
- Remove the old field before deploying any new consumers.
- Deploy the new producer alone and require all running clients to infer the rename.
- Introduce compatibility first, migrate consumers, and remove the old field afterward.
Explanation
Correct Answer
D:An expand-and-contract sequence gives both consumer versions a supported representation while they coexist. The old contract is retired only after its remaining users have migrated.
Incorrect Answers
A:That exposes still-running old consumers to an incompatible response during the overlap. Compatibility must exist before either side relies on the new-only contract.
B:Existing clients still require that field, so the first step breaks the active version before a replacement path is available.
C:A consumer's parser follows its implemented contract. It cannot be assumed to reinterpret an unfamiliar field without a compatible code change.
Question 111
An extractor’s quality declines after customers adopt a new document layout. What is the appropriate maintenance response?
- Assume the model has forgotten all earlier knowledge.
- Keep testing only the original layout.
- Analyze the new input slice and update evaluation coverage before revising the system.
- Keep the old evaluation dataset and assume production inputs have not changed.
Explanation
Correct Answer
C:The change in inputs supplies a concrete hypothesis. Targeted evidence can distinguish retrieval, prompting, and parsing failures.
Incorrect Answers
A:The observed layout change is a more specific cause than an unsupported universal model-memory claim.
B:That excludes the production condition associated with the regression.
D:A static test set can miss changes in the distribution of real inputs that explain the new production failures.
Question 112
A new configuration causes unauthorized tool attempts. What should an incident response prioritize?
- Contain the capability and restore a verified configuration.
- Increase autonomous retries.
- Retry every failed action immediately without disabling the faulty capability.
- Delete every trace before investigating.
Explanation
Correct Answer
A:Limiting exposure addresses the active failure while retained traces support diagnosis and a controlled recovery.
Incorrect Answers
B:More attempts can amplify the incident while the authorization defect remains.
C:Retries can repeat the harmful behavior while the incident remains active; containment must address the execution path itself.
D:Destroying evidence makes it harder to determine cause, affected actions, and remediation scope.
Question 113
A required model release is scheduled for retirement. What should maintenance include?
- Wait for production requests to fail before testing.
- Change a local model alias while leaving its provider mapping unchanged.
- Keep the existing model indefinitely by pinning its release identifier.
- Evaluate a supported replacement and migrate before the removal deadline.
Explanation
Correct Answer
D:A planned transition tests compatibility and quality while the existing system remains available for comparison.
Incorrect Answers
A:Reactive migration compresses validation into an outage and increases avoidable service disruption.
B:Renaming application configuration does not migrate traffic if requests still use the retiring provider identifier.
C:Pinning selects an exact release while it is available; it does not override a provider's documented retirement.
Question 114
Which TWO records best connect a deployed agent release to its validation? Choose TWO.
- A promise that future changes will be better.
- An unversioned screenshot of a successful chat.
- An immutable configuration/version manifest.
- A list of unrelated product features.
- Evaluation results associated with that same configuration.
Explanation
Correct Answers
C:The manifest identifies the actual components and settings that the deployment is intended to run.
E:Version-linked results show what was tested rather than relying on scores from an unrelated setup.
Incorrect Answers
A:A future intention provides no validation evidence for the current deployment.
B:A screenshot may omit configuration and cannot reliably identify the evaluated release.
D:Feature breadth does not show whether this release passed its acceptance criteria.
Question 115
A team must disable one experimental summarizer without stopping ticket intake. What design helps?
- A flag that hides the feature in the UI while its background jobs continue.
- A prompt telling customers not to use tickets.
- One global flag that terminates the whole application.
- A scoped feature flag around the optional summarization path.
Explanation
Correct Answer
D:Separating the optional path allows targeted disablement while the essential intake operation remains available.
Incorrect Answers
A:Hiding the entry point does not disable already running or independently scheduled backend behavior.
B:User instructions cannot replace an operational switch or preserve the intended service.
C:That removes the required intake capability along with the experimental feature.
Question 116
Users flag incorrect answers after deployment. What feedback record best supports improvement?
- Capture reviewed failure evidence with privacy controls.
- Retain all customer secrets for convenience.
- Store only the number of complaints.
- Replace the prompt after every unverified comment.
Explanation
Correct Answer
A:Reproducible, classified examples support targeted fixes and regression tests without collecting unnecessary sensitive data.
Incorrect Answers
B:Excessive sensitive collection is unnecessary for a useful, privacy-aware debugging record.
C:Counts show volume but cannot explain the conditions or mechanisms behind failures.
D:Uncontrolled changes can introduce regressions before the reported issue is understood.
Question 117
An API instruction must apply from the first turn. Where should it be supplied in a direct Messages request?
- The documented top-level system field.
- Only the browser’s page title.
- The API-key header.
- A file named CLAUDE.md that the API never receives.
Explanation
Correct Answer
A:This establishes initial system instructions. Model-specific support for later system messages does not replace the initial instruction mechanism.
Incorrect Answers
B:Interface metadata is not model input unless the application explicitly includes it.
C:Authentication fields establish service access, not natural-language behavioral instructions.
D:Direct model requests do not automatically read the application’s local repository instruction files.
Question 118
A supported request must call one particular tool rather than leave selection to Claude. What should the developer configure?
- Use automatic tool choice and mention the preferred tool in prose.
- The documented named-tool choice, subject to model compatibility.
- Use the any-tool setting when several tools are available.
- List the preferred tool first without setting a selection policy.
Explanation
Correct Answer
B:An explicit tool-selection control expresses the requirement more directly than relying only on suggestive prose.
Incorrect Answers
A:Automatic choice retains model discretion. A prose preference is weaker than the supported control requiring the named tool.
C:Requiring some tool does not select one particular tool among several. The requirement is for a named capability.
D:Definition order may influence selection but does not express the required named-tool constraint through the documented API control.
Question 119
A turn must produce a direct response without calling available tools. Which control is relevant where supported?
- Set max_tokens to zero as a tool policy.
- Set tool_choice to none.
- Delete the user’s request.
- Return a fake result for every tool.
Explanation
Correct Answer
B:This disables tool selection for the request while allowing a normal model response. This is an execution-selection control, distinct from merely asking for concise prose while leaving tool choice unconstrained.
Incorrect Answers
A:An output budget is not the appropriate control for selectively disabling tools while requesting a response.
C:Removing the task prevents a useful answer rather than controlling tool availability.
D:Fabricated observations are not needed to prevent invocation and would contaminate the conversation.
Question 120
With automatic tool choice, a request can be answered from supplied context. What behavior should the application permit?
- Invent a tool call to satisfy logging expectations.
- A direct response without any tool invocation.
- Run all declared tools regardless of the response.
- Reject every response lacking a tool call.
Explanation
Correct Answer
B:Automatic selection allows the model to decide whether a tool is needed; a tool definition does not require its use every turn.
Incorrect Answers
A:Logging should reflect actual behavior rather than manufacturing nonexistent invocations.
C:Availability is not an instruction to execute every capability.
D:That imposes a forced-tool requirement absent from automatic selection.