Question 200
A retrieval pipeline returns many neighboring pages that match keywords but not the intended question. Which implementation mechanism is needed?
- Retrieve relevant evidence per user question
- Use a task-specific worker for verbose analysis
- Filter irrelevant document chunks
- Put stable instructions ahead of volatile user material
Explanation
Correct Answer
C:Relevance reranking can reduce noise while keeping evidence that supports the answer, subject to this requirement: excessive filtering can discard essential exceptions or contradictory passages.
Incorrect Answers
A:Retrieval selects bounded passages pertinent to the current question and supplies their source context; this cannot substitute for the behavior that relevance reranking can reduce noise while keeping evidence that supports the answer, because selecting on superficial keywords alone may miss nuanced exceptions.
B:Delegating bulky examination to an isolated worker can keep main context focused; in contrast, relevance reranking can reduce noise while keeping evidence that supports the answer, while the worker's compressed report must disclose uncertainties and source references.
D:A stable prefix supports prompt-cache reuse and clearer separation of policy from per-request data; here, the controlling capability is that relevance reranking can reduce noise while keeping evidence that supports the answer, but if user text appears before the checkpoint it can invalidate shared cache hits.
Question 201
Before permitting broader unattended edits, a developer inspects effective privileges. Later, they need to isolate an unrelated incident investigation yet preserve progress on an unfinished refactor. Which THREE session and configuration actions fit? Choose THREE.
- Limit shell command authority in automation
- Start a new session for an unrelated objective
- Resume a relevant prior session intentionally
- Use hooks for deterministic lifecycle checks
- Use CLAUDE.md for repository instructions
- Review effective settings before broad automation
Explanation
Correct Answers
B:A clean context prevents old task details from biasing the new investigation; a crucial qualification is that shared persistent project guidance may still apply independently of session history.
C:Session resumption restores applicable prior conversational context for continuation; in production, resuming the wrong session can import stale assumptions or sensitive unrelated history.
F:Examining effective permission and hook configuration catches dangerous inherited allowances; the important constraint is that a printed prompt cannot override an enforced policy hierarchy.
Incorrect Answers
A:Restricted command/tool permissions reduce the possible effects of prompt errors or malicious code comments; here, the controlling capability is that a clean context prevents old task details from biasing the new investigation, but using a temporary checkout is helpful but does not itself revoke external credentials.
D:Configured hooks invoke developer-controlled checks at defined event boundaries; the needed capability is different—a clean context prevents old task details from biasing the new investigation—and a conversational instruction to perform the check is not equivalent to a guaranteed hook.
E:A project instruction file provides contextual guidance when Claude Code works in that repository, whereas the requirement calls for another behavior: a clean context prevents old task details from biasing the new investigation; it is explanatory context rather than a security boundary enforcing tool privileges.
Question 202
A public assistant is asked to generate instructions that could facilitate harmful misuse. Which technical control is appropriate?
- Sandbox uncertain code execution
- Canary a risky policy change
- Use a policy-based refusal and safe redirect
- Enforce least-privilege infrastructure permissions
Explanation
Correct Answer
C:A policy layer handles the request appropriately and communicates a safe boundary; a crucial qualification is that policy adherence should be evaluated; it is not guaranteed by a single warning prompt.
Incorrect Answers
A:Runtime isolation limits privileges, file access, and network effects of untrusted programs; this addresses a separate concern, not the fact that a policy layer handles the request appropriately and communicates a safe boundary; static text review cannot guarantee the code will behave safely.
B:A limited rollout measures refusal, error, and unauthorized action attempts before broad exposure; here, the controlling capability is that a policy layer handles the request appropriately and communicates a safe boundary, but immediate all-user changes make regressions harder to contain.
D:Backend credentials and IAM policies restrict it to the minimum necessary operations; by comparison, the intended solution relies on this: a policy layer handles the request appropriately and communicates a safe boundary; an unrestricted credential remains dangerous regardless of the model's stated intent.
Question 203
An HTTP client requests JSON but occasionally receives an error page from an intermediate proxy. Which of these alternatives serves the stated goal?
- Negotiate supported content types properly
- Track the API version contract
- Preserve future-compatible unknown fields cautiously
- Preserve UTF-8 content safely
Explanation
Correct Answer
A:Content-type checks help the parser distinguish valid API data from an unexpected gateway body; in production, blind JSON parsing obscures transport failures.
Incorrect Answers
B:Explicitly supported version settings and compatibility tests maintain a stable integration contract; the needed capability is different—content-type checks help the parser distinguish valid API data from an unexpected gateway body—and documentation examples for a different revision may not apply unchanged.
C:Tolerant read behavior can ignore unknown optional metadata while validating required fields, whereas the requirement calls for another behavior: content-type checks help the parser distinguish valid API data from an unexpected gateway body; silent acceptance of unknown privileged action types would be unsafe.
D:Proper encoding preserves characters consistently between client, API, and storage; however, the critical mechanism here is that content-type checks help the parser distinguish valid API data from an unexpected gateway body, and re-encoding bytes as the wrong character set can corrupt meaning.
Question 204
A support workflow must resume accurately after worker restarts or queue redelivery. What should the team do at this decision point?
- Store durable task state outside the model
- Keep a single direct model call
- Use idempotent side-effect handling
- Assign action limits in the agent harness
Explanation
Correct Answer
A:The application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage; in production, conversation context alone is not a transactional checkpoint or a durable job record.
Incorrect Answers
B:A direct request minimizes latency and operational state for a bounded answer; the needed capability is different—the application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage—and an agent loop would add failure modes without yielding a decision the model must make.
C:Stable operation identifiers and deduplication at the action service protect against duplicate writes, whereas the requirement calls for another behavior: the application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage; retrying a model-generated tool call blindly can repeat the business operation.
D:Host-controlled counters and deadlines bound execution regardless of what Claude decides next; however, the critical mechanism here is that the application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage, and prompt instructions alone cannot enforce a hard external quota.
Question 205
An agent cannot safely identify which tenant should receive a destructive configuration change. Which response is most defensible under these conditions?
- Finish with a user-facing answer after the last action
- Ask a clarifying question when critical input is missing
- Avoid splitting when coordination overhead dominates
- Observe a tool result before choosing the next action
Explanation
Correct Answer
B:Requesting an authorized scope prevents guesses from becoming unintended side effects; notably, an internal plan cannot legitimately invent the missing tenancy information.
Incorrect Answers
A:The model synthesizes a final response once no additional client tool request is needed; the requirement instead turns on this behavior: requesting an authorized scope prevents guesses from becoming unintended side effects; a raw tool result is not automatically a contextualized customer answer.
C:Keeping a bounded task local avoids orchestration latency and reconciliation errors; in contrast, requesting an authorized scope prevents guesses from becoming unintended side effects, while subagents should be justified by distinct work or isolation needs, not mere availability.
D:Alternating action and observation permits evidence-driven choices rather than a fully predetermined plan; this addresses a separate concern, not the fact that requesting an authorized scope prevents guesses from becoming unintended side effects; every loop must check for stop conditions and tool failures.
Question 206
An agent finishes three of five approved steps before the worker terminates unexpectedly. What should the team do at this decision point?
- Retry transient failures with backoff
- Record tool-call arguments and outcomes safely
- Persist failure checkpoints for resumability
- Tag each step with a correlation identifier
Explanation
Correct Answer
C:Checkpoint records allow a restart to continue without re-running completed side effects; in production, replaying the whole conversation may duplicate irreversible changes without idempotency controls.
Incorrect Answers
A:A bounded retry policy handles intermittent service errors without indefinitely hammering the dependency; here, the controlling capability is that checkpoint records allow a restart to continue without re-running completed side effects, but repeating a rejected unauthorized request cannot repair missing permission.
B:Structured traces capture call names, sanitized inputs, responses, timing, and error classification; the missing capability is more specific: checkpoint records allow a restart to continue without re-running completed side effects; logging must redact secrets and sensitive payload fields before persistence.
D:Propagating a stable trace ID across calls permits cause-and-effect analysis across services; the needed capability is different—checkpoint records allow a restart to continue without re-running completed side effects—and relying on timestamps alone makes concurrent sessions difficult to reconstruct.
Question 207
A finance assistant explains a recently amended regulation and proposes an interest percentage that must remain within a legally permitted range. Which TWO safeguards separately address freshness of evidence and correctness of numeric output? Choose TWO.
- Treat output as generated prediction, not a database read
- Use deterministic validation after generation
- Reserve enough output tokens for the desired result
- Use a computation tool for exact arithmetic
- Use retrieval for current factual evidence
Explanation
Correct Answers
B:Business code checks numeric bounds independently of the model's prose; fluent explanations cannot override a failed validation rule.
E:A retrieval step supplies updated source text before answer generation—with this limitation: retrieved content still needs authenticity and relevance checks.
Incorrect Answers
A:The language model generates plausible text rather than querying a company's live inventory by itself; by comparison, the intended solution relies on this: business code checks numeric bounds independently of the model's prose; it needs authorized retrieval to ground fast-changing private facts.
C:The configured generation maximum must be consistent with the required output length, whereas the requirement calls for another behavior: business code checks numeric bounds independently of the model's prose; raising it arbitrarily can increase possible cost without improving content accuracy.
D:A trusted arithmetic implementation calculates the numeric result instead of relying on next-token prediction; however, the critical mechanism here is that business code checks numeric bounds independently of the model's prose, and the model can explain the calculation but should not be the authoritative calculator.
Question 208
A developer adds a large system prompt, conversation history, tool results, and a final user message. Which engineering practice is the strongest match?
- Prune irrelevant historical turns when near capacity
- Count the complete request against context limits
- Handle context overflow before submission
- Reserve enough output tokens for the desired result
Explanation
Correct Answer
B:The context budget encompasses the relevant input contents and model-specific output allowances; a crucial qualification is that counting only the latest user text understates true token pressure.
Incorrect Answers
A:Context selection removes stale content while preserving state and key constraints; here, the controlling capability is that the context budget encompasses the relevant input contents and model-specific output allowances, but blind truncation from the end can delete vital recent instructions.
C:Chunking, retrieval, or compaction reduces input size while retaining decision-relevant evidence; by comparison, the intended solution relies on this: the context budget encompasses the relevant input contents and model-specific output allowances; increasing output max_tokens cannot make an overlong input fit.
D:The configured generation maximum must be consistent with the required output length; the needed capability is different—the context budget encompasses the relevant input contents and model-specific output allowances—and raising it arbitrarily can increase possible cost without improving content accuracy.
Question 209
A developer needs an agent to inspect files, edit code, and run local tests under project permissions. Which implementation mechanism is needed?
- Use Claude Desktop for an interactive desktop workspace
- Use CLAUDE.md for repository guidance in Claude Code
- Use Claude Code for repository-scoped coding assistance
- Use claude.ai for end-user conversational interaction
Explanation
Correct Answer
C:Claude Code operates in a coding environment with its own tools and repository context; the important constraint is that a generic text completion endpoint does not automatically manipulate local files.
Incorrect Answers
A:A desktop application manages interactive conversations and integrations under that client's configuration; the needed capability is different—claude Code operates in a coding environment with its own tools and repository context—and a hosted API request is not the same user-interface session.
B:Repository instruction files provide persistent project context to the Claude Code environment, whereas the requirement calls for another behavior: claude Code operates in a coding environment with its own tools and repository context; they are not automatically loaded into unrelated Messages API clients.
D:The hosted browser experience provides a ready-made conversational interface; the requirement instead turns on this behavior: claude Code operates in a coding environment with its own tools and repository context; it should not be assumed to inherit custom production app settings or session state.
Question 210
An extraction task can return unknown dates and the program must distinguish unavailable from blank strings. Which production safeguard is most relevant?
- Return actionable validation errors to callers
- Validate the outer JSON object schema
- Reject unexpected fields in business actions
- Handle null and missing values distinctly
Explanation
Correct Answer
D:Type-aware validation preserves nullability and missing-field semantics for downstream logic—with this limitation: coercing every absent property to an empty value can distort business decisions.
Incorrect Answers
A:A structured 4xx response names the offending field without leaking internal implementation details; the needed capability is different—type-aware validation preserves nullability and missing-field semantics for downstream logic—and a generic 500 would misleadingly suggest a server outage.
B:Schema checking enforces required keys, types, and nested structures consistently, whereas the requirement calls for another behavior: type-aware validation preserves nullability and missing-field semantics for downstream logic; a parseable JSON string can still violate the endpoint's semantic contract.
C:Application validation controls the allowed property set before any database mutation; however, the critical mechanism here is that type-aware validation preserves nullability and missing-field semantics for downstream logic, and the model may produce syntactically valid extra keys that are unsafe to execute.