Question 170 An API responds with HTTP 429 and a Retry-After interval during bulk classification. Which retry pattern is most reliable? Convert each 429 response into a successful classification. Retry every request immediately in parallel. Disable all timeout and retry limits permanently. Respect the retry guidance and use bounded exponential backoff with jitter. Explanation Correct Answer D:Backoff reduces contention and avoids synchronized retry storms. Incorrect Answers A:A transport error is not a semantic result. B:That can amplify rate limiting. C:Unbounded waits can stall the pipeline.
Question 171 A search tool returns multiple matching contracts but gives only unnamed strings. What output design most supports downstream decisions? Randomize results with no stable identifiers. Return stable contract IDs, titles, match reasons, and source references in typed fields. Omit source information to save a few tokens. Return only a count of matches. Explanation Correct Answer B:Identity and provenance let the agent distinguish candidates without guessing. Incorrect Answers A:Nondeterministic ordering makes follow-up references unreliable. C:Downstream validation needs a link to the underlying contracts. D:A count cannot identify which contract the user meant.
Question 172 A CLI job needs a bounded list of code-review findings with severity and file path fields. How should the runner obtain reliable parsing? Treat a missing file path as authorization to edit all files. Use the supported JSON output/schema flags and validate required fields downstream. Take the assistant’s confidence as proof its JSON is valid. Parse arbitrary markdown headings and hope they never change. Explanation Correct Answer B:Structured output makes the result predictable for automation. Incorrect Answers A:Missing fields should trigger validation, not broader scope. C:Confidence does not satisfy an API schema. D:Free-form structure is brittle under prompt variation.
Question 173 A coding agent stops midway through a refactor. The next request must retain the same edits and tool conversation. What session operation is appropriate? Generate a random new session ID and assume it maps to the old history. Resume the stored session using its supported identifier and preserved workspace state. Ask the model to guess previous test results. Start a fresh session with only “continue” and no files. Explanation Correct Answer B:Resumption keeps the existing conversation and execution history rather than starting over. Incorrect Answers A:Identity must refer to the actual persisted session. C:Unrecorded state should not be fabricated. D:A new session may not know which edits or decisions were completed.
Question 174 An incident classifier is overfitted to one template and fails on short informal user messages. How should the few-shot set change? Duplicate the long formal template many times. Remove all class definitions from the prompt. Include representative informal, formal, and borderline examples with consistent labels. Include mislabeled examples to increase randomness. Explanation Correct Answer C:Input diversity helps demonstrate the same decision criterion across styles. Incorrect Answers A:More copies reinforce the existing narrow pattern. B:Examples alone may not communicate exceptions or safety constraints. D:Inconsistent labels can degrade prediction quality.
Question 175 A duplicate-payment protection service reports “operation already completed; idempotency key matched.” What should the agent infer? Tell the user both payments definitely failed. Reconcile the original operation result rather than issuing a fresh payment with a new key. Discard all payment history and restart checkout. Generate a new payment identifier to force the transfer again. Explanation Correct Answer B:A duplicate response may signal prior success and must not trigger another charge. Incorrect Answers A:The service explicitly indicates an earlier operation was completed. C:Existing transaction state is essential to safe recovery. D:That bypasses idempotency and risks double payment.
Question 176 A code assistant receives a root rule and a path-specific rule that disagree about test commands. What should the project do? Allow random selection of the command on every run. Remove all test commands from both files. Prefer whichever file has a longer filename. Resolve the policy conflict explicitly and narrow the more specific exception to applicable files. Explanation Correct Answer D:Ambiguous conflicting rules cause inconsistent execution and should not be left to guesswork. Incorrect Answers A:Random test execution cannot meet a stable quality gate. B:That loses necessary verification guidance. C:Filename length does not define a meaningful policy.
Question 177 A team places a refund approval check in a PostToolUse hook after the payment service has processed it. Why is this placement wrong? Post-tool hooks cannot parse numeric amounts. A stronger user message automatically rewinds payments. The check must occur before the irreversible refund side effect. Refund tools cannot be called by agents. Explanation Correct Answer C:Post-tool observation is too late to guarantee a prerequisite approval. Incorrect Answers A:The problem is timing, not an inherent inability to inspect values. B:Text instructions do not roll back an executed financial transfer. D:Agent tools can propose financial actions when authorized.
Question 178 A classifier must decide whether a customer consent statement is affirmative. It labels “I might agree later” as yes. Which examples help? Show only enthusiastic yes statements. Tell the model every mention of consent means yes. Remove the no class entirely. Include positive, negative, conditional, and deferred-consent phrases with explicit labels. Explanation Correct Answer D:Contrasting nearby cases teaches that future possibility is not current consent. Incorrect Answers A:Those do not define the boundary for conditional language. B:A keyword rule would perpetuate the error. C:The classifier still needs to represent nonconsent.
Question 179 An aggregator produces a fluent report but cannot identify which worker sourced one numerical finding. What should the workflow enforce? Assume the figure is valid if the report formatting passes. Remove provenance from every other figure to appear consistent. Retain claim-level provenance through each worker handoff, or mark the figure unverified. Assign the figure to the worker with the fastest response. Explanation Correct Answer C:A polished summary cannot repair lost lineage. Incorrect Answers A:Presentation quality does not prove measurement. B:That sacrifices valid evidence. D:Latency is unrelated to source.
Question 180 A user changes the request from “propose a patch” to “apply the approved patch” in the same workspace. What must be updated? The execution authorization and tool mode, while preserving the vetted plan. Permission to modify any other repository. A random new goal unrelated to the patch. The entire project identity and all file IDs. Explanation Correct Answer A:A clear approval can justify a controlled transition from planning to editing. Incorrect Answers B:Approval for one patch is not universal write authority. C:The approved scope should remain the objective. D:Mode changes do not require restarting the project.
