Question 30
A document processing review requires two independent, explicitly defined outcomes: the first workstream must debug where a request stalls between the web UI, queue, and Claude gateway; a second must recover from transient 429 responses without synchronized retry storms. Which TWO actions directly provide these capabilities? Choose TWO.
- Use explicit HTTP method semantics
- Use exponential backoff with jitter
- Correlate distributed traces across services
- Enforce concurrency limits
- Pin dependencies with lockfiles
Explanation
Correct Answers
B:Increasing randomized delays reduce contention and respect service throttling signals; this directly supports task 2 to recover from transient 429 responses without synchronized retry storms, not the separate task or tasks to debug where a request stalls between the web UI, queue, and Claude gateway; immediate identical retries can extend an overload event.
C:A trace identifier links timing and errors through multiple process boundaries; this directly supports task 1 to debug where a request stalls between the web UI, queue, and Claude gateway, not the separate task or tasks to recover from transient 429 responses without synchronized retry storms; local logs with unrelated IDs cannot reconstruct the critical path reliably.
Incorrect Answers
A:GET should retrieve resources while state changes use appropriate mutation methods and authorization checks; this implements the different outcome to prevent a supposedly read-only status lookup from changing a server-side record rather than the required tasks to recover from transient 429 responses without synchronized retry storms and to debug where a request stalls between the web UI, queue, and Claude gateway, and putting write behavior behind a read endpoint undermines predictable retry and caching semantics, so it is not a direct solution to any of the assigned workstreams.
D:A semaphore or bounded queue keeps in-flight work within measured capacity; this implements the different outcome to prevent bursts of document jobs from overwhelming rate limits and memory rather than the required tasks to recover from transient 429 responses without synchronized retry storms and to debug where a request stalls between the web UI, queue, and Claude gateway, and launching every request at once turns a throughput problem into throttling, so it is not a direct solution to any of the assigned workstreams.
E:Locked versions and hashes constrain drift between build environments; this implements the different outcome to reproduce a working SDK installation across staging and production rather than the required tasks to recover from transient 429 responses without synchronized retry storms and to debug where a request stalls between the web UI, queue, and Claude gateway, and unbounded package updates can introduce new behavior during a routine rebuild, so it is not a direct solution to any of the assigned workstreams.
Question 31
Two separate tickets in a tool integration backlog require different controls. One needs to require stricter approval for actions that change customer data; the other needs to prevent malformed or unauthorized input from executing a customer data query. Which implementation pair is correct?
- First component: Separate read tools from write tools. Second component: Validate generated arguments in application code.
- First component: Add an approval boundary before sensitive writes. Second component: Validate generated arguments in application code.
- First component: Separate read tools from write tools. Second component: Restrict tool network credentials.
- First component: Restrict tool network credentials. Second component: Add an approval boundary before sensitive writes.
Explanation
Correct Answer
A:The first proposal fits its assigned role because distinct interfaces simplify least-privilege grants and side-effect controls, while the second proposal fits its assigned role because the executor must check types, ranges, identity, and scope independently of the model; their respective acceptance criteria are to require stricter approval for actions that change customer data and to prevent malformed or unauthorized input from executing a customer data query.
Incorrect Answers
B:The first proposal serves the different purpose to prevent a helpful assistant from issuing refunds automatically instead of the assigned first role because authorization by a user or operator can be enforced before the executor commits changes, while the second proposal fits its assigned role because the executor must check types, ranges, identity, and scope independently of the model; their respective acceptance criteria are to require stricter approval for actions that change customer data and to prevent malformed or unauthorized input from executing a customer data query.
C:The first proposal fits its assigned role because distinct interfaces simplify least-privilege grants and side-effect controls, while the second proposal serves the different purpose to stop a directory lookup function from also gaining unrestricted database write access instead of the assigned second role because minimal capabilities reduce risk when generated arguments or retrieved text are hostile; their respective acceptance criteria are to require stricter approval for actions that change customer data and to prevent malformed or unauthorized input from executing a customer data query.
D:The first proposal serves the different purpose to stop a directory lookup function from also gaining unrestricted database write access instead of the assigned first role because minimal capabilities reduce risk when generated arguments or retrieved text are hostile, while the second proposal serves the different purpose to prevent a helpful assistant from issuing refunds automatically instead of the assigned second role because authorization by a user or operator can be enforced before the executor commits changes; their respective acceptance criteria are to require stricter approval for actions that change customer data and to prevent malformed or unauthorized input from executing a customer data query.
Question 32
A technical lead requests a solution that can estimate model-token input demand before launching a large processing job within an inference routing project. Which design should be chosen?
- Send the correct API authentication header
- Specify request deadlines in the caller
- Use the Messages count_tokens capability
- Use JSON-safe types in request bodies
- Separate transport failures from model refusals
Explanation
Correct Answer
C:Token counting helps check input budgets before generation and cost planning; this behavior directly supports the need to estimate model-token input demand before launching a large processing job. Word or byte count is not a reliable substitute across diverse content.
Incorrect Answers
A:The API validates the provided credential before accepting a request; this applies to the different goal to avoid unauthenticated model requests from a backend worker, not the requirement to estimate model-token input demand before launching a large processing job; embedding a secret in a user prompt does not authenticate an HTTP call.
B:When an application needs to keep a web request from hanging indefinitely if a network path stalls, a client timeout bounds waiting for transport and response completion; the current capability must instead estimate model-token input demand before launching a large processing job, while model generation settings alone cannot enforce a socket deadline.
D:This approach serves the separate need to avoid rejecting tool schemas because serialized values are invalid because structured request encoding enforces objects, arrays, strings, booleans, and numeric representations; it does not establish the outcome to estimate model-token input demand before launching a large processing job, and python object repr strings are not valid JSON contracts.
E:A transport error occurs before a valid application-level completion; refusals are model outcomes; despite being useful when a service must surface network timeouts differently from a completed response that declines a task, it is not a solution to estimate model-token input demand before launching a large processing job; treating both as one generic retry condition can loop endlessly.
Question 33
A production-readiness check for a Claude Code repository component requires the service to deny destructive commands regardless of what a generated plan proposes. Which approach passes this check?
- Run repository initialization to create project guidance
- Use directory-scoped instructions
- Connect an MCP server to Claude Code
- Configure permissions in Claude Code settings
Explanation
Correct Answer
D:Explicit allow and deny rules establish executable tool policy rather than advisory prose; this behavior directly supports the need to deny destructive commands regardless of what a generated plan proposes. A markdown warning can be ignored or misinterpreted by the agent.
Incorrect Answers
A:This approach serves the separate need to establish a discoverable set of build and test conventions for a new Claude Code workspace because repository initialization can generate or prepare project instruction material that the team subsequently reviews; it does not establish the outcome to deny destructive commands regardless of what a generated plan proposes, and the resulting instructions are not a substitute for checking the actual build commands.
B:More local instruction files help scope guidance to the appropriate subtree; despite being useful when a service must apply specialised rules only to a sensitive infrastructure folder, it is not a solution to deny destructive commands regardless of what a generated plan proposes; root guidance alone can be too broad when different modules have conflicting conventions.
C:For the alternate objective to expose approved project-specific actions without copying credentials into prompts, an MCP integration declares external capabilities and their transport while access remains controlled by the server; the feature here must deny destructive commands regardless of what a generated plan proposes, and a server connection does not automatically make every operation safe or authorized.
Question 34
In a backend integration project, acceptance test A checks whether a component can maintain multiple Claude entry points without drifting retry and logging behavior. Acceptance test B checks whether a different component can allow older application clients to keep working while a new extraction field is introduced. Which plan satisfies both tests?
- First component: Enforce concurrency limits. Second component: Correlate distributed traces across services.
- First component: Refactor duplicated request orchestration. Second component: Enforce concurrency limits.
- First component: Refactor duplicated request orchestration. Second component: Version APIs and response schemas.
- First component: Correlate distributed traces across services. Second component: Version APIs and response schemas.
Explanation
Correct Answer
C:The first proposal fits its assigned role because a shared, tested client layer centralizes cross-cutting concerns and reduces inconsistent fixes, while the second proposal fits its assigned role because compatibility-aware contracts and versioned behavior protect consumers from surprise breaking changes; their respective acceptance criteria are to maintain multiple Claude entry points without drifting retry and logging behavior and to allow older application clients to keep working while a new extraction field is introduced.
Incorrect Answers
A:The first proposal serves the different purpose to prevent bursts of document jobs from overwhelming rate limits and memory instead of the assigned first role because a semaphore or bounded queue keeps in-flight work within measured capacity, while the second proposal serves the different purpose to debug where a request stalls between the web UI, queue, and Claude gateway instead of the assigned second role because a trace identifier links timing and errors through multiple process boundaries; their respective acceptance criteria are to maintain multiple Claude entry points without drifting retry and logging behavior and to allow older application clients to keep working while a new extraction field is introduced.
B:The first proposal fits its assigned role because a shared, tested client layer centralizes cross-cutting concerns and reduces inconsistent fixes, while the second proposal serves the different purpose to prevent bursts of document jobs from overwhelming rate limits and memory instead of the assigned second role because a semaphore or bounded queue keeps in-flight work within measured capacity; their respective acceptance criteria are to maintain multiple Claude entry points without drifting retry and logging behavior and to allow older application clients to keep working while a new extraction field is introduced.
D:The first proposal serves the different purpose to debug where a request stalls between the web UI, queue, and Claude gateway instead of the assigned first role because a trace identifier links timing and errors through multiple process boundaries, while the second proposal fits its assigned role because compatibility-aware contracts and versioned behavior protect consumers from surprise breaking changes; their respective acceptance criteria are to maintain multiple Claude entry points without drifting retry and logging behavior and to allow older application clients to keep working while a new extraction field is introduced.
Question 35
After pilot feedback, a document processing group needs two independent improvements. The first is to explain service limits or tool outages without leaking stack traces or secrets; the second is to avoid assuming a Claude Code repository setting automatically affects a direct Messages API call. Which engineering proposal handles both?
- First component: Create one session per user conversation. Second component: Place durable behavior instructions in the system parameter.
- First component: Return user-friendly error states. Second component: Create one session per user conversation.
- First component: Return user-friendly error states. Second component: Test interface-specific instruction behavior.
- First component: Place durable behavior instructions in the system parameter. Second component: Test interface-specific instruction behavior.
Explanation
Correct Answer
C:The first proposal fits its assigned role because structured error mapping presents actionable feedback while retaining detailed diagnostics in protected logs, while the second proposal fits its assigned role because different Claude surfaces have different configuration and context mechanisms; an integration must supply its own instructions; their respective acceptance criteria are to explain service limits or tool outages without leaking stack traces or secrets and to avoid assuming a Claude Code repository setting automatically affects a direct Messages API call.
Incorrect Answers
A:The first proposal serves the different purpose to keep follow-up requests linked without exposing another tenant's messages instead of the assigned first role because session scoping ensures history and tool results belong to the correct principal, while the second proposal serves the different purpose to ensure an API assistant consistently uses a defined support role across requests instead of the assigned second role because system-level guidance supplies application-wide instruction context, separate from each user's task content; their respective acceptance criteria are to explain service limits or tool outages without leaking stack traces or secrets and to avoid assuming a Claude Code repository setting automatically affects a direct Messages API call.
B:The first proposal fits its assigned role because structured error mapping presents actionable feedback while retaining detailed diagnostics in protected logs, while the second proposal serves the different purpose to keep follow-up requests linked without exposing another tenant's messages instead of the assigned second role because session scoping ensures history and tool results belong to the correct principal; their respective acceptance criteria are to explain service limits or tool outages without leaking stack traces or secrets and to avoid assuming a Claude Code repository setting automatically affects a direct Messages API call.
D:The first proposal serves the different purpose to ensure an API assistant consistently uses a defined support role across requests instead of the assigned first role because system-level guidance supplies application-wide instruction context, separate from each user's task content, while the second proposal fits its assigned role because different Claude surfaces have different configuration and context mechanisms; an integration must supply its own instructions; their respective acceptance criteria are to explain service limits or tool outages without leaking stack traces or secrets and to avoid assuming a Claude Code repository setting automatically affects a direct Messages API call.
Question 36
A backend integration review requires two independent, explicitly defined outcomes: the first workstream must record complete streamed output and token usage accurately; a second must extract claims from a document while retaining its visual layout. Which TWO actions directly provide these capabilities? Choose TWO.
- Handle the tool_use stop_reason
- Use supported document or PDF input
- Handle the max_tokens stop_reason
- Handle message_stop and final usage events
- Handle a refusal stop_reason safely
Explanation
Correct Answers
B:Document-aware input lets a capable model consider page content and graphics when supported; this directly supports task 2 to extract claims from a document while retaining its visual layout, not the separate task or tasks to record complete streamed output and token usage accurately; pasting only extracted text can discard useful visual context.
D:The final stream events signal completion and deliver accumulated metadata beyond individual text fragments; this directly supports task 1 to record complete streamed output and token usage accurately, not the separate task or tasks to extract claims from a document while retaining its visual layout; stopping after the first content delta truncates responses.
Incorrect Answers
A:This stop reason means tool calls need to be executed and their results returned before synthesis; this implements the different outcome to continue the agent loop when Claude is waiting for external observations rather than the required tasks to record complete streamed output and token usage accurately and to extract claims from a document while retaining its visual layout, and terminating at this point strands an unfinished task, so it is not a direct solution to any of the assigned workstreams.
C:The finish reason identifies truncation, so a caller may need a continuation or a revised token budget; this implements the different outcome to recover a response that ended because its output budget was exhausted rather than the required tasks to record complete streamed output and token usage accurately and to extract claims from a document while retaining its visual layout, and treating that response as a complete answer can hide missing fields, so it is not a direct solution to any of the assigned workstreams.
E:A refusal is a deliberate non-completion path that requires a defined application response; this implements the different outcome to avoid parsing safety refusals as valid JSON business records rather than the required tasks to record complete streamed output and token usage accurately and to extract claims from a document while retaining its visual layout, and repeatedly increasing the token limit does not resolve a content refusal, so it is not a direct solution to any of the assigned workstreams.
Question 37
An engineer extending a retrieval pipeline workflow has been asked to fit a large knowledge base into the context of a customer answer. Which choice addresses the request most precisely?
- Prioritize high-signal retrieved passages
- Refresh stale evidence before reuse
- Reserve space for the model's output
- Separate long-term memory from active scratchpad
- Deduplicate overlapping passages
Explanation
Correct Answer
A:Ranking by relevance and authority places the most useful evidence ahead of incidental matches; this behavior directly supports the need to fit a large knowledge base into the context of a customer answer. Dumping every document into the prompt can obscure the actual answer.
Incorrect Answers
B:When an application needs to avoid answering a pricing question from last month's retrieved cache, freshness checks determine whether a saved passage remains authoritative for a dynamic fact; the current capability must instead fit a large knowledge base into the context of a customer answer, while high semantic similarity says nothing about timeliness.
C:Input assembly must leave room for the requested completion and any needed reasoning; the supported outcome is to avoid using the entire context budget on retrieved input before generating a response, whereas this scenario requires the ability to fit a large knowledge base into the context of a customer answer; a technically accepted long prompt can still prevent a useful-length answer.
D:This approach serves the separate need to prevent outdated user preferences from being treated as current task instructions because explicit storage and freshness rules govern what is retained and when it is reintroduced; it does not establish the outcome to fit a large knowledge base into the context of a customer answer, and an unbounded persistent transcript can accumulate contradictory assumptions.
E:Overlap detection reduces redundant context without throwing away unique clauses; despite being useful when a service must save tokens when several retrieval chunks contain the same repeated policy section, it is not a solution to fit a large knowledge base into the context of a customer answer; simply increasing retrieval count may provide many copies of the same material.
Question 38
A tool integration review asks the team to close two unrelated gaps without merging their execution paths. Service A must enable a customer service agent to prepare but not autonomously approve account closures; service B must perform open-ended troubleshooting where tool selection changes after each observation. Which paired remediation works?
- First component: Publish an MCP server for reusable integrations. Second component: Use an autonomous agent for exploratory work.
- First component: Use system or repository instructions for stable behavior. Second component: Publish an MCP server for reusable integrations.
- First component: Combine tool access with human confirmation. Second component: Use an autonomous agent for exploratory work.
- First component: Combine tool access with human confirmation. Second component: Use system or repository instructions for stable behavior.
Explanation
Correct Answer
C:The first proposal fits its assigned role because tools supply actionable capabilities while a separate approval step governs sensitive execution, while the second proposal fits its assigned role because model-directed planning supports adaptive sequences when the next action is not known in advance; their respective acceptance criteria are to enable a customer service agent to prepare but not autonomously approve account closures and to perform open-ended troubleshooting where tool selection changes after each observation.
Incorrect Answers
A:The first proposal serves the different purpose to make one internal ticketing connector available to several agent clients instead of the assigned first role because MCP standardizes tool and resource discovery across compatible applications, while the second proposal fits its assigned role because model-directed planning supports adaptive sequences when the next action is not known in advance; their respective acceptance criteria are to enable a customer service agent to prepare but not autonomously approve account closures and to perform open-ended troubleshooting where tool selection changes after each observation.
B:The first proposal serves the different purpose to make a coding or API assistant follow organization-wide conventions instead of the assigned first role because instructions shape model behavior and task framing without executing external operations, while the second proposal serves the different purpose to make one internal ticketing connector available to several agent clients instead of the assigned second role because MCP standardizes tool and resource discovery across compatible applications; their respective acceptance criteria are to enable a customer service agent to prepare but not autonomously approve account closures and to perform open-ended troubleshooting where tool selection changes after each observation.
D:The first proposal fits its assigned role because tools supply actionable capabilities while a separate approval step governs sensitive execution, while the second proposal serves the different purpose to make a coding or API assistant follow organization-wide conventions instead of the assigned second role because instructions shape model behavior and task framing without executing external operations; their respective acceptance criteria are to enable a customer service agent to prepare but not autonomously approve account closures and to perform open-ended troubleshooting where tool selection changes after each observation.
Question 39
A technical lead requests a solution that can decide which customer fields must be redacted before model requests within a production API project. Which design should be chosen?
- Record nonfunctional latency targets
- Classify data sensitivity up front
- Calculate expected traffic and concurrency
- Define failure and recovery objectives
- Identify authoritative source systems
Explanation
Correct Answer
B:Data classification guides minimization, access controls, retention, and acceptable transfers; this behavior directly supports the need to decide which customer fields must be redacted before model requests. Sanitizing only after a model has seen secret input is too late.
Incorrect Answers
A:A measurable response-time objective constrains interaction design and infrastructure choices; despite being useful when a service must decide between streaming and offline processing for a time-sensitive support assistant, it is not a solution to decide which customer fields must be redacted before model requests; functional prompts alone cannot establish acceptable performance.
C:Workload assumptions drive rate-limit planning, queue capacity, and retry strategy; this applies to the different goal to provision an interactive application for peak seasonal traffic, not the requirement to decide which customer fields must be redacted before model requests; average request volume alone hides burst behavior.
D:When an application needs to keep a regulated triage service usable during an upstream API outage, availability, recovery time, and fallback behavior must be specified as measurable operational needs; the current capability must instead decide which customer fields must be redacted before model requests, while a happy-path prototype misses behavior under dependency failures.
E:Source-of-truth mapping identifies which verified systems must supply facts and what freshness guarantees apply; the supported outcome is to prevent a Claude assistant from inventing account balances when it lacks current ledger data, whereas this scenario requires the ability to decide which customer fields must be redacted before model requests; model inference is not a substitute for a transactional database.
Question 40
In a customer platform project, acceptance test A checks whether a component can avoid parsing safety refusals as valid JSON business records. Acceptance test B checks whether a different component can preserve a customer's follow-up reference to an earlier case number. Which plan satisfies both tests?
- First component: Handle a refusal stop_reason safely. Second component: Handle the max_tokens stop_reason.
- First component: Handle a refusal stop_reason safely. Second component: Supply previous turns on a new API request.
- First component: Supply a tool definition with input schema. Second component: Supply previous turns on a new API request.
- First component: Handle the max_tokens stop_reason. Second component: Supply a tool definition with input schema.
Explanation
Correct Answer
B:The first proposal fits its assigned role because a refusal is a deliberate non-completion path that requires a defined application response, while the second proposal fits its assigned role because stateless requests need relevant history passed back in the messages sequence; their respective acceptance criteria are to avoid parsing safety refusals as valid JSON business records and to preserve a customer's follow-up reference to an earlier case number.
Incorrect Answers
A:The first proposal fits its assigned role because a refusal is a deliberate non-completion path that requires a defined application response, while the second proposal serves the different purpose to recover a response that ended because its output budget was exhausted instead of the assigned second role because the finish reason identifies truncation, so a caller may need a continuation or a revised token budget; their respective acceptance criteria are to avoid parsing safety refusals as valid JSON business records and to preserve a customer's follow-up reference to an earlier case number.
C:The first proposal serves the different purpose to let Claude request a read-only inventory lookup with typed arguments instead of the assigned first role because a declared tool name, description, and schema constrain the interface offered to the model, while the second proposal fits its assigned role because stateless requests need relevant history passed back in the messages sequence; their respective acceptance criteria are to avoid parsing safety refusals as valid JSON business records and to preserve a customer's follow-up reference to an earlier case number.
D:The first proposal serves the different purpose to recover a response that ended because its output budget was exhausted instead of the assigned first role because the finish reason identifies truncation, so a caller may need a continuation or a revised token budget, while the second proposal serves the different purpose to let Claude request a read-only inventory lookup with typed arguments instead of the assigned second role because a declared tool name, description, and schema constrain the interface offered to the model; their respective acceptance criteria are to avoid parsing safety refusals as valid JSON business records and to preserve a customer's follow-up reference to an earlier case number.