Question 90
A failing external dependency causes the entire app to build up pending requests and exhaust workers. What would best address the stated failure mode?
- Use a sandbox backend for tool integration tests
- Use a safe degraded response on dependency outage
- Use a circuit breaker for cascading 5xx failures
- Return actionable validation errors to callers
Explanation
Correct Answer
C:A circuit breaker temporarily blocks repeated failing calls while permitting controlled recovery probes; in production, it should not be used to hide permanent programming errors.
Incorrect Answers
A:A nonproduction environment exercises authorization, payload mapping, and failures safely; here, the controlling capability is that a circuit breaker temporarily blocks repeated failing calls while permitting controlled recovery probes, but faking every backend call can hide actual integration incompatibilities.
B:A fallback can queue the request or return a limited verified response without fabricating model results; the missing capability is more specific: a circuit breaker temporarily blocks repeated failing calls while permitting controlled recovery probes; retrying indefinitely inside the user's HTTP connection harms availability.
D:A structured 4xx response names the offending field without leaking internal implementation details; the needed capability is different—a circuit breaker temporarily blocks repeated failing calls while permitting controlled recovery probes—and a generic 500 would misleadingly suggest a server outage.
Question 91
A status update supports only draft, pending, approved, or rejected transitions. How should this requirement be met?
- Constrain nested object shapes
- Represent allowed values as enums where appropriate
- Provide precise tool descriptions and argument types
- Use an Anthropic server-executed tool where supported
Explanation
Correct Answer
B:Constrained value sets prevent unsupported strings from reaching backend business logic—with this limitation: valid enum membership alone cannot authorize an illegal state transition.
Incorrect Answers
A:Structured nested validation rejects incomplete or malformed records consistently; however, the critical mechanism here is that constrained value sets prevent unsupported strings from reaching backend business logic, and accepting arbitrary object shapes can produce unpredictable writes.
C:Clear names, schema fields, and descriptions help the model choose and populate the intended operation; this cannot substitute for the behavior that constrained value sets prevent unsupported strings from reaching backend business logic, because tool schemas still need server-side validation of all inputs.
D:Server tools execute on the service side and return associated results during the model interaction; in contrast, constrained value sets prevent unsupported strings from reaching backend business logic, while they have distinct data and capability boundaries from the customer's custom functions.
Question 92
An API request must target a Claude model made available to the account and region. What is the best fit for the described situation?
- Choose cache TTL based on reuse interval
- Set an explicit max_tokens budget
- Choose a supported model identifier
- Supply alternating conversation turns in messages
Explanation
Correct Answer
C:The model field identifies the deployed model; production should verify current IDs and compatibility; in production, guessing an informal nickname can yield an invalid-model error or unexpected alias changes.
Incorrect Answers
A:A supported longer-duration cache may suit reuse that occurs after the default window; here, the controlling capability is that the model field identifies the deployed model; production should verify current IDs and compatibility, but longer cache writes can cost more, so economics depend on actual hit frequency.
B:The request parameter limits output generation independently from the size of the prompt; the missing capability is more specific: the model field identifies the deployed model; production should verify current IDs and compatibility; a larger max_tokens value does not expand the model's input context window.
D:Request messages carry explicit user and assistant roles because the API does not automatically retain prior chat state; the needed capability is different—the model field identifies the deployed model; production should verify current IDs and compatibility—and omitting history makes follow-up references ambiguous unless reintroduced elsewhere.
Question 93
A finished marketing draft should not influence an unrelated software incident analysis. Which action most directly closes the gap?
- Summarize intermediate reasoning and preserve outcomes
- Start a new clean window after task boundary
- Reassert the current task objective concisely
- Retain exact quoted requirements when necessary
Explanation
Correct Answer
B:Separating independent sessions avoids accidental carryover of assumptions and private context; the new session must still receive any relevant persistent project policy.
Incorrect Answers
A:Compaction keeps conclusions, evidence, and open questions while discarding low-value intermediate tokens; by comparison, the intended solution relies on this: separating independent sessions avoids accidental carryover of assumptions and private context; overaggressive summarization can remove the details that make conclusions defensible.
C:Reintroducing the active goal and constraints guides continuation back to the approved scope, whereas the requirement calls for another behavior: separating independent sessions avoids accidental carryover of assumptions and private context; repetition cannot override the need to preserve factual state accurately.
D:Exact critical terms should remain intact as source data rather than be loosely paraphrased; however, the critical mechanism here is that separating independent sessions avoids accidental carryover of assumptions and private context, and a small numerical wording change can reverse a regulatory decision.
Question 94
A long case review must carry established facts and unresolved issues into a later stage without full transcripts. What is the most direct implementation choice?
- Keep pointers to source documents
- Retrieve relevant evidence per user question
- Isolate unrelated subtasks in separate contexts
- Keep a structured conversation summary
Explanation
Correct Answer
D:A summary stores salient decisions, constraints, and references in a reduced representation; a crucial qualification is that compression may omit nuance and needs checking for critical obligations.
Incorrect Answers
A:References and identifiers enable later retrieval without copying the entire source into every prompt; in contrast, a summary stores salient decisions, constraints, and references in a reduced representation, while pointers alone do not make an inaccessible document available at inference time.
B:Retrieval selects bounded passages pertinent to the current question and supplies their source context; this addresses a separate concern, not the fact that a summary stores salient decisions, constraints, and references in a reduced representation; selecting on superficial keywords alone may miss nuanced exceptions.
C:Distinct subagent contexts prevent irrelevant data from filling the coordinator's main conversation; here, the controlling capability is that a summary stores salient decisions, constraints, and references in a reduced representation, but true confidentiality still depends on tool access and data-sharing controls.
Question 95
A reverse proxy returned an HTML gateway error rather than the expected JSON message. Which action most directly closes the gap?
- Set the requested API version header where required
- Use typed deserialization for a structured result
- Use POST for a Messages generation request
- Parse an HTTP status before decoding success payload
Explanation
Correct Answer
D:The caller distinguishes transport/application failure from successful responses before parsing their bodies; attempting to deserialize every response as a Claude message conceals gateway faults.
Incorrect Answers
A:Version negotiation constrains accepted fields and response behavior for an endpoint; here, the controlling capability is that the caller distinguishes transport/application failure from successful responses before parsing their bodies, but omitting required version information or assuming an arbitrary revision can break compatibility.
B:Typed parsing converts validated JSON into an explicit application data contract; the missing capability is more specific: the caller distinguishes transport/application failure from successful responses before parsing their bodies; free-form text parsing with ad hoc regular expressions is brittle under wording changes.
C:The endpoint creates a new inference result from a JSON request and is invoked as an HTTP POST; by comparison, the intended solution relies on this: the caller distinguishes transport/application failure from successful responses before parsing their bodies; treating it as a cacheable read-only GET would not express the required request body.
Question 96
A support agent reads a third-party article that orders it to expose a secret; an incoming customer prompt also attempts to trigger an unauthorized shipment lookup. Which TWO defenses apply at the separate content-trust and tool-authorization boundaries? Choose TWO.
- Authorize each requested operation at the resource level
- Verify tool permissions independently of text
- Filter outgoing sensitive data at the egress layer
- Isolate risky browsing from privileged tools
- Treat a retrieved webpage as untrusted data
Explanation
Correct Answers
B:Server-side authorization checks bind each requested resource to the authenticated principal—with this limitation: a system-prompt ban on disclosure cannot replace real access control.
E:External content is evidence, not an instruction source allowed to override the application's policies, subject to this requirement: prompt labeling helps but tool-side access restrictions must remain enforced.
Incorrect Answers
A:Access policies check whether that principal may perform the exact requested change on the target object; however, the critical mechanism here is that server-side authorization checks bind each requested resource to the authenticated principal, and successful login alone does not grant permission across all records.
C:Deterministic output and action checks block unauthorized exfiltration before it leaves the trusted environment; this cannot substitute for the behavior that server-side authorization checks bind each requested resource to the authenticated principal, because model self-policing alone cannot guarantee detection of every malicious instruction.
D:Separate trust zones and scoped workers prevent public content from directly wielding private capabilities; in contrast, server-side authorization checks bind each requested resource to the authenticated principal, while isolation fails if credentials or tool results leak back across the boundary.
Question 97
A support team needs repeatable procedures for API timeouts, credential rotation, and degraded service. Which option best fits the described workflow?
- Write a design review before implementation
- Use automated integration tests before release
- Maintain a change log with impact scope
- Maintain operations runbooks
Explanation
Correct Answer
D:Operational documentation specifies symptoms, escalation, mitigations, and safe recovery steps; a crucial qualification is that a developer's local notebook is not a reliable on-call procedure.
Incorrect Answers
A:Architecture review assesses interfaces, failure modes, and control ownership before code changes; in contrast, operational documentation specifies symptoms, escalation, mitigations, and safe recovery steps, while implementing immediately can lock in unsafe assumptions and costly redesign.
B:Integration tests check how independently coded components actually interact under realistic conditions; this addresses a separate concern, not the fact that operational documentation specifies symptoms, escalation, mitigations, and safe recovery steps; unit tests of isolated helpers cannot reveal all cross-component contract failures.
C:Release records state compatibility, affected endpoints, and migration actions; here, the controlling capability is that operational documentation specifies symptoms, escalation, mitigations, and safe recovery steps, but a model-generated summary alone does not define the actual deployed contract.
Question 98
A technical design prompt combines goals, constraints, evidence, and output format. Which production safeguard is most relevant?
- Separate requirements from source data
- Use named sections for long complex requirements
- Avoid unnecessary verbosity constraints
- Use explicit delimiters for source passages
Explanation
Correct Answer
B:Structured headings make independent instructions easier to distinguish and check—with this limitation: structure cannot rescue contradictory or factually impossible requirements.
Incorrect Answers
A:Structural boundaries distinguish data to analyze from commands the assistant should follow; however, the critical mechanism here is that structured headings make independent instructions easier to distinguish and check, and external content should not gain authority merely because it appears near instructions.
C:Output guidance should match the amount of reasoning and explanation the task actually needs; this cannot substitute for the behavior that structured headings make independent instructions easier to distinguish and check, because longer responses can add unsupported details rather than increasing precision.
D:Marking the passage's bounds clarifies which content is data rather than a user command; in contrast, structured headings make independent instructions easier to distinguish and check, while delimiters help interpretation but do not replace tool-level security.
Question 99
A prompt works on three simple English messages but fails on long multilingual customer inputs. Which development approach should be selected?
- Validate prompt variables before interpolation
- Test prompt examples for representativeness
- Include a counterexample with correction
- Provide contrastive edge-case examples
Explanation
Correct Answer
B:Broader cases reveal where examples bias the model toward a narrow pattern; the important constraint is that more examples do not automatically fix poor dataset coverage.
Incorrect Answers
A:Input checks constrain length, types, and permitted resource scopes before prompt assembly, whereas the requirement calls for another behavior: broader cases reveal where examples bias the model toward a narrow pattern; merely surrounding an unsafe value with quotation marks does not establish authorization.
C:Showing an unacceptable output and its corrected form highlights a specific formatting mistake; the requirement instead turns on this behavior: broader cases reveal where examples bias the model toward a narrow pattern; a negative example without the correct alternative may create confusion.
D:Distinct positive and near-miss cases teach the actual boundary between labels; this cannot substitute for the behavior that broader cases reveal where examples bias the model toward a narrow pattern, because repeating many easy cases offers little information about ambiguous ones.
Question 100
An autonomous triage job must stop before it exceeds its approved tool-call and cost budget. What is the appropriate next design choice?
- Store durable task state outside the model
- Assign action limits in the agent harness
- Use a model-directed agent loop
- Keep model context limited to relevant evidence
Explanation
Correct Answer
B:Host-controlled counters and deadlines bound execution regardless of what Claude decides next; a crucial qualification is that prompt instructions alone cannot enforce a hard external quota.
Incorrect Answers
A:The application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage; here, the controlling capability is that host-controlled counters and deadlines bound execution regardless of what Claude decides next, but conversation context alone is not a transactional checkpoint or a durable job record.
C:Claude selects an action after observing each result, with the application enforcing iteration and permission limits; by comparison, the intended solution relies on this: host-controlled counters and deadlines bound execution regardless of what Claude decides next; the autonomy adds variability and demands hard budgets, logs, and a stopping policy.
D:A curated evidence window preserves useful facts while discarding noisy intermediate transcripts; the needed capability is different—host-controlled counters and deadlines bound execution regardless of what Claude decides next—and pruning must preserve commitments and results required for downstream steps.