Question 380
Clients encounter throttling and the server communicates a suggested pause before retrying. Which action most directly closes the gap?
- Measure both request and token rates
- Respect retry-after guidance when available
- Use retry-aware HTTP error handling
- Queue overload rather than dropping accepted work
Explanation
Correct Answer
B:Retry scheduling honors the provider's feedback and avoids synchronized hammering; the important constraint is that relying only on fixed zero-delay retries is counterproductive.
Incorrect Answers
A:Capacity planning accounts for relevant request, input, and output token budgets, whereas the requirement calls for another behavior: retry scheduling honors the provider's feedback and avoids synchronized hammering; counting only completed HTTP calls can miss token-based constraints.
C:Status classification drives different responses for transient and permanent errors; the requirement instead turns on this behavior: retry scheduling honors the provider's feedback and avoids synchronized hammering; blindly retrying invalid JSON does not improve correctness.
D:A persistent queue buffers approved tasks with controlled worker concurrency; this cannot substitute for the behavior that retry scheduling honors the provider's feedback and avoids synchronized hammering, because unbounded in-process arrays risk memory exhaustion and data loss.
Question 381
A multi-stage fulfillment assistant must keep a stable shipment reference, distinguish missing records from forbidden access, and avoid flooding the model with unused product fields. Which THREE tool-result design practices meet these needs? Choose THREE.
- Return only the fields necessary for a decision
- Limit untrusted text in tool outputs
- Preserve tool error codes separately from text
- Include stable identifiers in tool results
- Mark truly required arguments in the schema
- Report tool failures explicitly
Explanation
Correct Answers
A:Structured filtering reduces context load and unnecessary data disclosure; the implementation must still recognize that the model should not receive personal or unrelated fields merely because they exist upstream.
C:Typed errors guide safe retries, clarification, or escalation without inventing data; in production, free-form error prose can be misread or stripped by downstream clients.
D:IDs let downstream operations refer unambiguously to retrieved business objects; human-readable names alone may collide across tenants or records.
Incorrect Answers
B:Treating tool output as low-trust evidence and trimming unnecessary text reduces injection surfaces; the missing capability is more specific: structured filtering reduces context load and unnecessary data disclosure; a tool result is not an application-level system command.
E:The schema identifies mandatory fields so calls missing them can be rejected before dispatch, whereas the requirement calls for another behavior: structured filtering reduces context load and unnecessary data disclosure; optional property defaults should not hide missing business-critical identifiers.
F:A structured error result communicates the failed operation and supports safe recovery; however, the critical mechanism here is that structured filtering reduces context load and unnecessary data disclosure, and silently replacing an error with an empty success encourages false completion claims.
Question 382
A greeting normalizer is taking disproportionately long because advanced reasoning is always enabled. Which change would satisfy the constraint?
- Use extra reasoning on high-complexity tasks
- Use low-temperature settings when supported for consistency
- Use a lighter reasoning configuration for trivial tasks
- Verify model thinking compatibility
Explanation
Correct Answer
C:Lower effort on supported models is appropriate for a narrowly specified routine transformation; notably, under-allocation can hurt difficult exceptions, so evaluate the change.
Incorrect Answers
A:Supported thinking or effort settings allocate greater reasoning capacity to difficult deliberation; however, the critical mechanism here is that lower effort on supported models is appropriate for a narrowly specified routine transformation, and added reasoning can increase latency and cost and still requires factual validation.
B:Lower randomness can reduce sampling variability in supported configurations; the requirement instead turns on this behavior: lower effort on supported models is appropriate for a narrowly specified routine transformation; it does not guarantee deterministic output or eliminate model errors.
D:The integrator checks supported thinking parameters for the chosen version and request mode; in contrast, lower effort on supported models is appropriate for a narrowly specified routine transformation, while feature names and modes can differ between model releases.
Question 383
An assistant makes twenty small tool-assisted inference calls to finish one support case. Which of these alternatives serves the stated goal?
- Recognize latency due to extra reasoning
- Count the complete request against context limits
- Handle output truncation explicitly
- Bound total cost across agent turns
Explanation
Correct Answer
D:Aggregating tokens and actions by logical task reveals overall expenditure and quota needs; the important constraint is that per-call estimates overlook the cumulative work of multi-turn orchestration.
Incorrect Answers
A:Reasoning controls and task complexity affect total latency independently of answer length; by comparison, the intended solution relies on this: aggregating tokens and actions by logical task reveals overall expenditure and quota needs; reducing displayed text alone may not remove the underlying thinking cost.
B:The context budget encompasses the relevant input contents and model-specific output allowances; the needed capability is different—aggregating tokens and actions by logical task reveals overall expenditure and quota needs—and counting only the latest user text understates true token pressure.
C:Stop-reason inspection catches the partial response before parsing or persistence, whereas the requirement calls for another behavior: aggregating tokens and actions by logical task reveals overall expenditure and quota needs; an arbitrary string closure should not be treated as a validated complete JSON object.
Question 384
Build pipelines require credentials but must not check them into source control. Which approach targets the deciding condition?
- Protect API keys through deployment secrets
- Pin dependency versions for reproducible builds
- Use code review on tool schema changes
- Redact credentials before logging
Explanation
Correct Answer
A:Secret injection provides runtime access without publishing keys in code or artifacts, although obscuring a key in a code comment is not adequate credential protection.
Incorrect Answers
B:A lockfile and tested upgrade process preserve a known dependency graph; however, the critical mechanism here is that secret injection provides runtime access without publishing keys in code or artifacts, and unrestricted installs are vulnerable to unreviewed breaking changes.
C:Reviewing the schema and authorization path catches privilege expansion before release; the requirement instead turns on this behavior: secret injection provides runtime access without publishing keys in code or artifacts; merely changing the tool description does not guarantee safe backend behavior.
D:Logging middleware removes secrets before storage or export to observability systems; this cannot substitute for the behavior that secret injection provides runtime access without publishing keys in code or artifacts, because restricting dashboard viewers does not erase sensitive fields already written.
Question 385
A new summarization prompt changes outputs for thousands of users and needs controlled exposure. Which approach best meets this requirement?
- Conduct a staged production rollout
- Maintain operations runbooks
- Categorize and triage production incidents
- Write a design review before implementation
Explanation
Correct Answer
A:Canary releases compare a limited population against baseline before broad enablement, subject to this requirement: immediate full deployment magnifies regressions and complicates rollback.
Incorrect Answers
B:Operational documentation specifies symptoms, escalation, mitigations, and safe recovery steps; here, the controlling capability is that canary releases compare a limited population against baseline before broad enablement, but a developer's local notebook is not a reliable on-call procedure.
C:Incident classification routes problems to the correct service, integration, or prompt team; the missing capability is more specific: canary releases compare a limited population against baseline before broad enablement; treating every failure as bad model reasoning conceals infrastructure issues.
D:Architecture review assesses interfaces, failure modes, and control ownership before code changes; by comparison, the intended solution relies on this: canary releases compare a limited population against baseline before broad enablement; implementing immediately can lock in unsafe assumptions and costly redesign.
Question 386
A repository assistant must reject destructive shell commands before execution and capture the result of permitted commands afterward. Which TWO deterministic Claude Code hook stages address those separate moments? Choose TWO.
- Use tool allowlists as a baseline defense
- Use a post-tool hook for outcome auditing
- Use a mandatory human approval mechanism
- Use a pre-tool hook for prohibited actions
Explanation
Correct Answers
B:A lifecycle hook can capture sanitized execution results and support monitoring or follow-up checks—with this limitation: a post-action hook cannot retroactively prevent a harmful command that already ran; the post-action record complements rather than replaces the separate pre-execution prohibition.
D:A pre-execution hook can inspect and block the operation before it reaches the shell; its operational boundary is that the hook runs outside the model's judgment, but its own policy must be correctly implemented; the command must be stopped before execution, while the separate post-action step supplies evidence.
Incorrect Answers
A:Tool availability and credential permissions restrict the reachable attack surface by design; however, the critical mechanism here is that a lifecycle hook can capture sanitized execution results and support monitoring or follow-up checks, and hooks are additional safeguards, not replacements for minimizing available power.
C:An enforced approval boundary stops execution until confirmation is provided; this cannot substitute for the behavior that a lifecycle hook can capture sanitized execution results and support monitoring or follow-up checks, because a model's promise to ask first is not itself a technical approval gate; human consent does not automatically create an event log after each permitted command.
Question 387
The lead agent must answer a legal comparison while delegating source collection to a narrowly scoped helper. Which change would satisfy the constraint?
- Delegate retrieval to a research worker
- Use a summarizing worker for context compression
- Delegate quality checking to a critic worker
- Keep model context limited to relevant evidence
Explanation
Correct Answer
A:A research worker gathers references and returns a compact evidence set to the lead; the important constraint is that it should not be allowed to publish the final legal conclusion as an independent authority.
Incorrect Answers
B:A dedicated summarizer condenses retained facts with provenance and open issues for the parent context; the requirement instead turns on this behavior: a research worker gathers references and returns a compact evidence set to the lead; compression may omit nuance unless the system checks critical commitments and references.
C:An evaluator examines the artifact against explicit tests instead of composing the primary answer; this cannot substitute for the behavior that a research worker gathers references and returns a compact evidence set to the lead, because critique must have actionable criteria and should not be equated with guaranteed correctness.
D:A curated evidence window preserves useful facts while discarding noisy intermediate transcripts; in contrast, a research worker gathers references and returns a compact evidence set to the lead, while pruning must preserve commitments and results required for downstream steps.
Question 388
A customer-service task only needs shipping status but the upstream record also includes identity documents. How should the implementation handle this need?
- Apply retention and deletion rules to transcripts
- Minimize personal fields before inference
- Redact secrets from observability logs
- Authenticate the caller before processing account actions
Explanation
Correct Answer
B:Data minimization removes unrelated personal information prior to sending content to the model; its operational boundary is that a request to disregard a field still exposes that field if it is transmitted.
Incorrect Answers
A:Storage controls govern how long records, logs, and derived artifacts remain available; this cannot substitute for the behavior that data minimization removes unrelated personal information prior to sending content to the model, because a model's context window does not define legal data retention requirements.
C:Redaction removes sensitive fields before events reach metrics and log sinks; this addresses a separate concern, not the fact that data minimization removes unrelated personal information prior to sending content to the model; encrypting logs does not justify logging unnecessary secrets indefinitely.
D:Identity verification establishes who is requesting the action before tool access is considered; here, the controlling capability is that data minimization removes unrelated personal information prior to sending content to the model, but a convincing statement of ownership inside a prompt is not authentication.
Question 389
A custom front end needs an ongoing bidirectional signaling channel between browser and its own backend. Which design would meet this acceptance criterion?
- Use connection reuse for frequent calls
- Use HTTP request-response for discrete inference
- Use server-sent events for streamed responses
- Use WebSockets for application-managed duplex traffic
Explanation
Correct Answer
D:A WebSocket supports messages in both directions on one persistent connection; the important constraint is that its availability does not imply the standard Claude Messages endpoint is a WebSocket API.
Incorrect Answers
A:Persistent connections reduce repeated connection setup overhead; by comparison, the intended solution relies on this: a WebSocket supports messages in both directions on one persistent connection; pool sizing still has to respect concurrency and upstream rate limits.
B:Standard HTTPS requests fit isolated generation operations without long-lived bidirectional connections; the needed capability is different—a WebSocket supports messages in both directions on one persistent connection—and opening a persistent socket adds state and lifecycle complexity where it is unnecessary.
C:SSE carries incremental events from the service to the client as generation proceeds, whereas the requirement calls for another behavior: a WebSocket supports messages in both directions on one persistent connection; it is not a general-purpose bidirectional client messaging transport.
Question 390
A new model version has changed pricing or supported context characteristics relative to the old one. Which technical control is appropriate?
- Update cost and context assumptions
- Run a canary rollout for a new model version
- Revalidate tool schemas and outcomes
- Use a fast tier for request classification
Explanation
Correct Answer
A:Recalculating budgets avoids relying on stale token or feature assumptions; a crucial qualification is that product names alone do not prove pricing or capacity parity.
Incorrect Answers
B:Limited exposure allows observation of quality, refusal, latency, and tool usage before full migration; by comparison, the intended solution relies on this: recalculating budgets avoids relying on stale token or feature assumptions; a model upgrade can affect the entire user experience even when code is unchanged.
C:Integration tests inspect requested tool names, arguments, and final answer handling; the needed capability is different—recalculating budgets avoids relying on stale token or feature assumptions—and syntax compatibility does not guarantee identical agent decisions.
D:A small capable classifier avoids spending top-tier reasoning cost on simple routing, whereas the requirement calls for another behavior: recalculating budgets avoids relying on stale token or feature assumptions; poor routing accuracy can still misdirect expensive follow-up steps.