Question 440
A coding assistant needs a reusable integration exposed by a separate MCP server. What is the right implementation response?
- Use a deterministic application function without model dispatch
- Use a client-executed custom tool
- Include stable identifiers in tool results
- Use an MCP-provided tool through a configured client
Explanation
Correct Answer
D:The MCP client discovers the server's tools and invokes permitted operations through a standardized protocol, although MCP transport does not replace authentication or authorization of the business system.
Incorrect Answers
A:Conventional code runs the calculation directly and avoids unnecessary probabilistic tool selection; the missing capability is more specific: the MCP client discovers the server's tools and invokes permitted operations through a standardized protocol; a language-model intermediary would add cost without changing the arithmetic rules.
B:The API returns a tool_use request and the application performs the authorized operation itself; by comparison, the intended solution relies on this: the MCP client discovers the server's tools and invokes permitted operations through a standardized protocol; tool declarations do not grant the vendor access to private infrastructure.
C:IDs let downstream operations refer unambiguously to retrieved business objects; the needed capability is different—the MCP client discovers the server's tools and invokes permitted operations through a standardized protocol—and human-readable names alone may collide across tenants or records.
Question 441
A constrained runtime cannot install an SDK but can send secure HTTP requests with JSON payloads. What should be implemented to solve this problem?
- Use a raw HTTPS REST client when necessary
- Authenticate with a server-managed API credential
- Use a wrapper service for shared organizational controls
- Use an official SDK for conventional API calls
Explanation
Correct Answer
A:Direct REST requests can call the same documented endpoint with correct headers and body; notably, the developer must implement encoding, retries, and response parsing independently.
Incorrect Answers
B:A trusted backend injects and safeguards the credential when making requests; in contrast, direct REST requests can call the same documented endpoint with correct headers and body, while embedding the key in front-end JavaScript makes it available to site visitors.
C:A centralized proxy can enforce cross-cutting organizational controls before forwarding requests; this addresses a separate concern, not the fact that direct REST requests can call the same documented endpoint with correct headers and body; the proxy also becomes a reliability and credential-protection responsibility.
D:A supported client library wraps the underlying service API and common network details; here, the controlling capability is that direct REST requests can call the same documented endpoint with correct headers and body, but the SDK does not decide business-level retries or permissions automatically.
Question 442
Claude requests a valid tool call, but the application sees a 500 from its own invoice backend. Which development approach should be selected?
- Trace the external tool service
- Treat 5xx as potential upstream service faults
- Reproduce with a raw valid API request
- Inspect parsed model content separately
Explanation
Correct Answer
A:A distributed trace follows the tool dispatcher to the failing downstream dependency; its operational boundary is that rewriting system instructions will not repair an unavailable invoice endpoint.
Incorrect Answers
B:The client applies guarded retries and monitoring while preserving safe degraded behavior; this addresses a separate concern, not the fact that a distributed trace follows the tool dispatcher to the failing downstream dependency; treating an outage as a prompt defect misdirects investigation.
C:An isolated direct call distinguishes transport and wrapper behavior from a more complex integration path; here, the controlling capability is that a distributed trace follows the tool dispatcher to the failing downstream dependency, but a successful API response does not automatically validate downstream application logic.
D:Output-level checks isolate semantic and schema failures from network transport errors; the missing capability is more specific: a distributed trace follows the tool dispatcher to the failing downstream dependency; treating every wrong answer as an API outage conceals prompt and evaluation problems.
Question 443
A service repeatedly sends the same lengthy instruction block and reference manual to Claude. Which change would satisfy the constraint?
- Inspect cache read and creation usage
- Cache a stable shared prompt prefix
- Correlate batch results by custom identifiers
- Choose cache TTL based on reuse interval
Explanation
Correct Answer
B:Prompt caching can reuse an identical eligible prefix instead of reprocessing it fully on every call; in production, changing text or tool definitions ahead of the checkpoint can invalidate prefix reuse.
Incorrect Answers
A:Cache usage metrics distinguish newly written content from tokens read from cache; the missing capability is more specific: prompt caching can reuse an identical eligible prefix instead of reprocessing it fully on every call; an assumed hit without usage evidence can hide prompt-prefix changes.
C:Stable request IDs let the application map each result to the correct original record; the needed capability is different—prompt caching can reuse an identical eligible prefix instead of reprocessing it fully on every call—and relying on result ordering can mix customer records or output destinations.
D:A supported longer-duration cache may suit reuse that occurs after the default window, whereas the requirement calls for another behavior: prompt caching can reuse an identical eligible prefix instead of reprocessing it fully on every call; longer cache writes can cost more, so economics depend on actual hit frequency.
Question 444
A service has lost access to a required private tool after an organizational policy change. How should this requirement be met?
- Escalate a persistent authorization denial
- Use a retry budget for transient calls
- Treat 401/403 responses as credential or access problems
- Apply safe fallback when retries exhaust
Explanation
Correct Answer
A:Operators investigate credentials and permissions through the proper access workflow; a crucial qualification is that retries or prompt pressure must not be used to bypass access controls.
Incorrect Answers
B:A bounded attempt count and deadline balance recovery against cost and responsiveness; by comparison, the intended solution relies on this: operators investigate credentials and permissions through the proper access workflow; endless retries can exhaust quotas and mask permanent faults.
C:Authentication and entitlement checks distinguish invalid credentials from forbidden resource access; the needed capability is different—operators investigate credentials and permissions through the proper access workflow—and changing prompt wording cannot grant missing API permission.
D:Returning a clear incomplete status or known safe alternative avoids false success claims, whereas the requirement calls for another behavior: operators investigate credentials and permissions through the proper access workflow; inventing an answer to preserve a green status is operationally misleading.
Question 445
A help-desk endpoint only needs a short answer from supplied, verified FAQ context with no tools or follow-ups. Which solution should the engineer adopt?
- Keep a single direct model call
- Use an evaluator-optimizer loop
- Select a sequential prompt chain
- Store durable task state outside the model
Explanation
Correct Answer
A:A direct request minimizes latency and operational state for a bounded answer—with this limitation: an agent loop would add failure modes without yielding a decision the model must make.
Incorrect Answers
B:A grader produces targeted feedback that the writer consumes on the next bounded iteration; this cannot substitute for the behavior that a direct request minimizes latency and operational state for a bounded answer, because without a maximum attempts policy the system can burn tokens without improving the draft.
C:The first output becomes the next step's input in a deliberate dependency chain; in contrast, a direct request minimizes latency and operational state for a bounded answer, while running these stages concurrently would deny the reviewer the draft it must inspect.
D:The application persists task IDs, stage progress, completed actions, and retry metadata in controlled storage; this addresses a separate concern, not the fact that a direct request minimizes latency and operational state for a bounded answer; conversation context alone is not a transactional checkpoint or a durable job record.
Question 446
A production legal extractor needs explicit approval before its underlying model changes. Which production safeguard is most relevant?
- Retain a tested rollback version
- Store environment-specific endpoints in configuration
- Validate configuration at startup
- Pin tested model identifiers for controlled services
Explanation
Correct Answer
D:A fixed supported model ID avoids unexpected alias migrations during the approval window, although pinning cannot prevent a provider from eventually retiring a version.
Incorrect Answers
A:Keeping the previously tested artifact enables a measured rollback when safe; the missing capability is more specific: a fixed supported model ID avoids unexpected alias migrations during the approval window; a rollback should consider state migrations and outstanding jobs.
B:Deployment configuration injects environment values without rewriting application logic; by comparison, the intended solution relies on this: a fixed supported model ID avoids unexpected alias migrations during the approval window; hardcoded production URLs increase the chance of accidental test writes.
C:Early checks fail fast with a clear operator error before the service accepts work; the needed capability is different—a fixed supported model ID avoids unexpected alias migrations during the approval window—and deferring discovery to live traffic increases incident impact.
Question 447
A staff member wants to ask Claude general questions in a browser without building a custom integration. Which response is most defensible under these conditions?
- Use Claude Desktop for an interactive desktop workspace
- Use claude.ai for end-user conversational interaction
- Use a direct API integration for a branded customer feature
- Use CLAUDE.md for repository guidance in Claude Code
Explanation
Correct Answer
B:The hosted browser experience provides a ready-made conversational interface; a crucial qualification is that it should not be assumed to inherit custom production app settings or session state.
Incorrect Answers
A:A desktop application manages interactive conversations and integrations under that client's configuration; here, the controlling capability is that the hosted browser experience provides a ready-made conversational interface, but a hosted API request is not the same user-interface session.
C:The Messages API allows application code to own product behavior and its user experience; by comparison, the intended solution relies on this: the hosted browser experience provides a ready-made conversational interface; consumer chat sessions are not an app's programmable backend contract.
D:Repository instruction files provide persistent project context to the Claude Code environment; the needed capability is different—the hosted browser experience provides a ready-made conversational interface—and they are not automatically loaded into unrelated Messages API clients.
Question 448
A new deployment introduces intermittent 400s, 429s, and 5xx failures at different rates. Which action most directly closes the gap?
- Return actionable validation errors to callers
- Redact credentials before logging
- Record latency per integration stage
- Count errors by class and endpoint
Explanation
Correct Answer
D:Structured metrics reveal whether the issue is client validation, quota pressure, or upstream availability; a single error counter masks very different remediation paths.
Incorrect Answers
A:A structured 4xx response names the offending field without leaking internal implementation details; here, the controlling capability is that structured metrics reveal whether the issue is client validation, quota pressure, or upstream availability, but a generic 500 would misleadingly suggest a server outage.
B:Logging middleware removes secrets before storage or export to observability systems; the missing capability is more specific: structured metrics reveal whether the issue is client validation, quota pressure, or upstream availability; restricting dashboard viewers does not erase sensitive fields already written.
C:Timed spans separate request phases and identify the bottleneck for targeted optimization; by comparison, the intended solution relies on this: structured metrics reveal whether the issue is client validation, quota pressure, or upstream availability; total elapsed time alone cannot localize the slow dependency.
Question 449
A primary assistant needs only an answer summary from a very large source comparison. Which production safeguard is most relevant?
- Use targeted tool requests instead of bulk dumps
- Filter irrelevant document chunks
- Use a task-specific worker for verbose analysis
- Reassert the current task objective concisely
Explanation
Correct Answer
C:Delegating bulky examination to an isolated worker can keep main context focused; notably, the worker's compressed report must disclose uncertainties and source references.
Incorrect Answers
A:Narrow queries reduce context consumption and sensitive data exposure simultaneously; however, the critical mechanism here is that delegating bulky examination to an isolated worker can keep main context focused, and returned results still need authorization before the query executes.
B:Relevance reranking can reduce noise while keeping evidence that supports the answer; the requirement instead turns on this behavior: delegating bulky examination to an isolated worker can keep main context focused; excessive filtering can discard essential exceptions or contradictory passages.
D:Reintroducing the active goal and constraints guides continuation back to the approved scope; in contrast, delegating bulky examination to an isolated worker can keep main context focused, while repetition cannot override the need to preserve factual state accurately.
Question 450
A synthesized research note may be challenged and auditors must reopen the original records. What change is best aligned with the goal?
- Summarize intermediate reasoning and preserve outcomes
- Keep pointers to source documents
- Keep a structured conversation summary
- Start a new clean window after task boundary
Explanation
Correct Answer
B:References and identifiers enable later retrieval without copying the entire source into every prompt, although pointers alone do not make an inaccessible document available at inference time.
Incorrect Answers
A:Compaction keeps conclusions, evidence, and open questions while discarding low-value intermediate tokens; the needed capability is different—references and identifiers enable later retrieval without copying the entire source into every prompt—and overaggressive summarization can remove the details that make conclusions defensible.
C:A summary stores salient decisions, constraints, and references in a reduced representation; however, the critical mechanism here is that references and identifiers enable later retrieval without copying the entire source into every prompt, and compression may omit nuance and needs checking for critical obligations.
D:Separating independent sessions avoids accidental carryover of assumptions and private context; the requirement instead turns on this behavior: references and identifiers enable later retrieval without copying the entire source into every prompt; the new session must still receive any relevant persistent project policy.