Question 230
Tests for an after-sales issue resolver will pass only if an order adjustment uses the current order record and if a delegated reviewer has no production write capability. Which proposed design directly satisfies both checks?
- require the order lookup result before calculating or applying a change; and give the reviewer every administrative tool and ask for restraint
- require the order lookup result before calculating or applying a change; and provide only the read-oriented tools required by that reviewer
- issue the update using details inferred only from conversation; and provide only the read-oriented tools required by that reviewer
- issue the update using details inferred only from conversation; and give the reviewer every administrative tool and ask for restraint
Explanation
Correct Answer
B:In an after-sales issue resolver, the dependent operation needs authoritative state and must not guess from the customer's description; alongside this, agent-scoped tool exposure prevents a write path in a read-only role, satisfying the distinct requirements that an order adjustment uses the current order record and that a delegated reviewer has no production write capability.
Incorrect Answers
A:In an after-sales issue resolver, the dependent operation needs authoritative state and must not guess from the customer's description, but prompt restraint is weaker than withholding execution capability; therefore the separate requirement that a delegated reviewer has no production write capability would remain unsatisfied.
C:In an after-sales issue resolver, agent-scoped tool exposure prevents a write path in a read-only role, but unverified user text can reference the wrong order or outdated state; therefore the separate requirement that an order adjustment uses the current order record would remain unsatisfied.
D:In an after-sales issue resolver, unverified user text can reference the wrong order or outdated state and prompt restraint is weaker than withholding execution capability; the design would fail both conditions: an order adjustment uses the current order record and a delegated reviewer has no production write capability.
Question 231
For a mobile client bug-fix task, a reviewer identifies two independent obligations: the implementation plan receives stakeholder sign-off and a broad tool catalog does not overwhelm a small task. Which proposal should be approved?
- present ordered changes, risks, and verification steps before switching to execution; and load all available MCP tools into every specialist turn
- treat the agent's internally generated plan as implicit approval; and load all available MCP tools into every specialist turn
- treat the agent's internally generated plan as implicit approval; and supply only context-relevant tools for the current agent
- present ordered changes, risks, and verification steps before switching to execution; and supply only context-relevant tools for the current agent
Explanation
Correct Answer
D:In a mobile client bug-fix task, an explicit review point aligns the agent's work with the authorized scope; alongside this, reducing irrelevant tool definitions improves selection and saves context, satisfying the distinct requirements that the implementation plan receives stakeholder sign-off and that a broad tool catalog does not overwhelm a small task.
Incorrect Answers
A:In a mobile client bug-fix task, an explicit review point aligns the agent's work with the authorized scope, but tool bloat makes the available actions harder to discriminate; therefore the separate requirement that a broad tool catalog does not overwhelm a small task would remain unsatisfied.
B:In a mobile client bug-fix task, the system cannot infer external authorization from its own reasoning and tool bloat makes the available actions harder to discriminate; the design would fail both conditions: the implementation plan receives stakeholder sign-off and a broad tool catalog does not overwhelm a small task.
C:In a mobile client bug-fix task, reducing irrelevant tool definitions improves selection and saves context, but the system cannot infer external authorization from its own reasoning; therefore the separate requirement that the implementation plan receives stakeholder sign-off would remain unsatisfied.
Question 232
Production readiness for a quarterly market-intelligence coordinator depends on ensuring that tool output is normalized before it enters the agent context and that multiple subagent outputs can be traced to their individual tasks. Select the most appropriate combined approach.
- preprocess only the original user message; and attach task IDs and source references to returned artifacts
- use a PostToolUse processing hook to standardize the returned data; and strip origin metadata from each response before merging
- preprocess only the original user message; and strip origin metadata from each response before merging
- use a PostToolUse processing hook to standardize the returned data; and attach task IDs and source references to returned artifacts
Explanation
Correct Answer
D:In a quarterly market-intelligence coordinator, post-execution interception can clean result fields and attach useful status without changing the model's tool selection; alongside this, traceable results let the coordinator identify which delegation produced each claim, satisfying the distinct requirements that tool output is normalized before it enters the agent context and that multiple subagent outputs can be traced to their individual tasks.
Incorrect Answers
A:In a quarterly market-intelligence coordinator, traceable results let the coordinator identify which delegation produced each claim, but input normalization does not correct inconsistent tool outputs; therefore the separate requirement that tool output is normalized before it enters the agent context would remain unsatisfied.
B:In a quarterly market-intelligence coordinator, post-execution interception can clean result fields and attach useful status without changing the model's tool selection, but source-less aggregation impairs diagnosis and fact checking; therefore the separate requirement that multiple subagent outputs can be traced to their individual tasks would remain unsatisfied.
C:In a quarterly market-intelligence coordinator, input normalization does not correct inconsistent tool outputs and source-less aggregation impairs diagnosis and fact checking; the design would fail both conditions: tool output is normalized before it enters the agent context and multiple subagent outputs can be traced to their individual tasks.
Question 233
Before expanding a merge-request policy gate, the owner must demonstrate that an automation expects a defined JSON result schema, while also demonstrating that a shared skill can take a supplied target name or file path. Which design provides both controls?
- ask for valid JSON in prose but accept any text result; and copy a separate static instruction file for every target path
- ask for valid JSON in prose but accept any text result; and define how the skill accepts and uses invocation arguments
- use the CLI structured-output schema option for the response; and copy a separate static instruction file for every target path
- use the CLI structured-output schema option for the response; and define how the skill accepts and uses invocation arguments
Explanation
Correct Answer
D:In a merge-request policy gate, schema-constrained output is more robust for downstream integrations than prompt-only formatting; alongside this, parameterized workflows apply one reviewed procedure to different resources without hardcoding targets, satisfying the distinct requirements that an automation expects a defined JSON result schema and that a shared skill can take a supplied target name or file path.
Incorrect Answers
A:In a merge-request policy gate, a natural-language instruction does not guarantee parseable or complete fields and duplicated procedures are costly to maintain and can diverge; the design would fail both conditions: an automation expects a defined JSON result schema and a shared skill can take a supplied target name or file path.
B:In a merge-request policy gate, parameterized workflows apply one reviewed procedure to different resources without hardcoding targets, but a natural-language instruction does not guarantee parseable or complete fields; therefore the separate requirement that an automation expects a defined JSON result schema would remain unsatisfied.
C:In a merge-request policy gate, schema-constrained output is more robust for downstream integrations than prompt-only formatting, but duplicated procedures are costly to maintain and can diverge; therefore the separate requirement that a shared skill can take a supplied target name or file path would remain unsatisfied.
Question 234
For a test failure triage bot, acceptance depends on demonstrating that a model learns a boundary between superficially similar cases. Which approach is the strongest fit?
- show only easy positive samples
- audit examples against the current decision rubric before use
- include a demonstration showing how an absent field is represented
- include contrasting positive and negative demonstrations
Explanation
Correct Answer
D:For a test failure triage bot, near-boundary examples communicate the decisive distinction and reduce false positives; this directly establishes that a model learns a boundary between superficially similar cases instead of relying on an unrelated workflow control.
Incorrect Answers
A:In a test failure triage bot, the model may generalize the label to unrelated but similar-looking items; as a result, the stated requirement that a model learns a boundary between superficially similar cases would still not be met.
B:Although consistent labels reinforce the intended criteria rather than training ambiguity into the prompt, that only supports the distinct goal that few-shot material does not contain contradictory answer labels, not the requirement that a model learns a boundary between superficially similar cases in a test failure triage bot.
C:Although explicit missing-value examples clarify that omissions are not permission to hallucinate, that only supports the distinct goal that an unusual null-value case is handled without inventing data, not the requirement that a model learns a boundary between superficially similar cases in a test failure triage bot.
Question 235
The release gate for a claims form processing assistant requires that source identity survives multi-agent research handoffs; it also requires that a conclusion supported by model agreement is also tested against source evidence. Which TWO actions meet the specified checks? Choose TWO.
- strip all citation fields before sending the writer its inputs
- carry citation identifiers and retrieval context through synthesis
- require source and business-rule checks regardless of model agreement
- promote agreement alone to a guaranteed correctness signal
Explanation
Correct Answers
B:For a claims form processing assistant, provenance metadata must remain attached when intermediate summaries are merged, which directly establishes that source identity survives multi-agent research handoffs; the separate requirement that a conclusion supported by model agreement is also tested against source evidence is handled by a different action.
C:For a claims form processing assistant, correlated model errors can survive consensus without external verification, which directly establishes that a conclusion supported by model agreement is also tested against source evidence; the separate requirement that source identity survives multi-agent research handoffs is handled by a different action.
Incorrect Answers
A:For a claims form processing assistant, the writer cannot reliably reconstruct which source supports each conclusion, so this option cannot establish that source identity survives multi-agent research handoffs; correctly addressing a conclusion supported by model agreement is also tested against source evidence elsewhere does not repair that deficiency.
D:For a claims form processing assistant, shared biases can produce identical wrong conclusions, so this option cannot establish that a conclusion supported by model agreement is also tested against source evidence; correctly addressing source identity survives multi-agent research handoffs elsewhere does not repair that deficiency.
Question 236
The release gate for a policy document normalization service requires that a specialist handoff retains its prerequisites and evidence; it also requires that client-side tool requests trigger another model turn after execution. Which TWO actions meet the specified checks? Choose TWO.
- end the loop whenever assistant text appears
- handoff only a conversational sentence saying the stage is finished
- send a structured state packet with completed checks and unresolved decisions
- branch on stop_reason tool_use and execute returned tool_use blocks
Explanation
Correct Answers
C:For a policy document normalization service, a typed transition preserves obligations and facts between workflow stages, which directly establishes that a specialist handoff retains its prerequisites and evidence; the separate requirement that client-side tool requests trigger another model turn after execution is handled by a different action.
D:For a policy document normalization service, the response explicitly signals pending client execution and the loop must provide results before another model decision, which directly establishes that client-side tool requests trigger another model turn after execution; the separate requirement that a specialist handoff retains its prerequisites and evidence is handled by a different action.
Incorrect Answers
A:For a policy document normalization service, text may accompany a tool request and is not the completion signal, so this option cannot establish that client-side tool requests trigger another model turn after execution; correctly addressing a specialist handoff retains its prerequisites and evidence elsewhere does not repair that deficiency.
B:For a policy document normalization service, the receiving stage cannot audit what actually passed, so this option cannot establish that a specialist handoff retains its prerequisites and evidence; correctly addressing client-side tool requests trigger another model turn after execution elsewhere does not repair that deficiency.
Question 237
In a merge-request policy gate, the acceptance criteria require that a CI job invokes Claude without a terminal conversation and that an approved plan is translated into controlled edits and tests. Which paired implementation meets both?
- run Claude Code in non-interactive print mode with -p; and remain in read-only planning and claim the code was updated
- run Claude Code in non-interactive print mode with -p; and exit the planning stage and perform scoped implementation with verification
- launch an interactive Claude terminal in an unattended runner; and remain in read-only planning and claim the code was updated
- launch an interactive Claude terminal in an unattended runner; and exit the planning stage and perform scoped implementation with verification
Explanation
Correct Answer
B:In a merge-request policy gate, headless invocation permits scripted automation and exits after the requested task; alongside this, planning describes intended operations while execution applies and checks them, satisfying the distinct requirements that a CI job invokes Claude without a terminal conversation and that an approved plan is translated into controlled edits and tests.
Incorrect Answers
A:In a merge-request policy gate, headless invocation permits scripted automation and exits after the requested task, but a proposal alone does not modify repository state; therefore the separate requirement that an approved plan is translated into controlled edits and tests would remain unsatisfied.
C:In a merge-request policy gate, an interactive session can block automation waiting for a person and a proposal alone does not modify repository state; the design would fail both conditions: a CI job invokes Claude without a terminal conversation and an approved plan is translated into controlled edits and tests.
D:In a merge-request policy gate, planning describes intended operations while execution applies and checks them, but an interactive session can block automation waiting for a person; therefore the separate requirement that a CI job invokes Claude without a terminal conversation would remain unsatisfied.
Question 238
In a technical review of an account recovery support process, the architect must address the following constraints together: an ordinary in-policy question is not escalated needlessly; the system distinguishes acceptable and unacceptable edge cases. Which change set is most defensible?
- send every low-risk question to a human as a default; and specify inclusion criteria and explicit exclusion conditions
- send every low-risk question to a human as a default; and provide only a broad topic label as the classification rule
- reserve human handoff for specified triggers and let normal cases proceed; and provide only a broad topic label as the classification rule
- reserve human handoff for specified triggers and let normal cases proceed; and specify inclusion criteria and explicit exclusion conditions
Explanation
Correct Answer
D:In an account recovery support process, well-defined boundaries maintain autonomy without bypassing risky decisions; alongside this, negative boundaries teach the model when a near-match must not trigger, satisfying the distinct requirements that an ordinary in-policy question is not escalated needlessly and that the system distinguishes acceptable and unacceptable edge cases.
Incorrect Answers
A:In an account recovery support process, negative boundaries teach the model when a near-match must not trigger, but blanket escalation defeats the purpose of routine assistance; therefore the separate requirement that an ordinary in-policy question is not escalated needlessly would remain unsatisfied.
B:In an account recovery support process, blanket escalation defeats the purpose of routine assistance and generic labels leave ambiguous boundary cases undefined; the design would fail both conditions: an ordinary in-policy question is not escalated needlessly and the system distinguishes acceptable and unacceptable edge cases.
C:In an account recovery support process, well-defined boundaries maintain autonomy without bypassing risky decisions, but generic labels leave ambiguous boundary cases undefined; therefore the separate requirement that the system distinguishes acceptable and unacceptable edge cases would remain unsatisfied.
Question 239
For a structured case-file ingestion workflow, a reviewer identifies two independent obligations: a formatter handles diverse legitimate examples consistently and a missing value is represented without fabricating a string. Which proposal should be approved?
- give a small set of representative input-output demonstrations; and force the model to populate every unknown value with a guess
- give a small set of representative input-output demonstrations; and allow null in the field schema and define its semantics
- supply only a bare instruction to keep formatting consistent; and allow null in the field schema and define its semantics
- supply only a bare instruction to keep formatting consistent; and force the model to populate every unknown value with a guess
Explanation
Correct Answer
B:In a structured case-file ingestion workflow, few-shot examples show the intended mapping more concretely than broad style adjectives; alongside this, explicit nullable types distinguish genuine absence from empty or invented data, satisfying the distinct requirements that a formatter handles diverse legitimate examples consistently and that a missing value is represented without fabricating a string.
Incorrect Answers
A:In a structured case-file ingestion workflow, few-shot examples show the intended mapping more concretely than broad style adjectives, but fabricated values preserve schema appearance but destroy factual accuracy; therefore the separate requirement that a missing value is represented without fabricating a string would remain unsatisfied.
C:In a structured case-file ingestion workflow, explicit nullable types distinguish genuine absence from empty or invented data, but abstract instructions leave many output choices open; therefore the separate requirement that a formatter handles diverse legitimate examples consistently would remain unsatisfied.
D:In a structured case-file ingestion workflow, abstract instructions leave many output choices open and fabricated values preserve schema appearance but destroy factual accuracy; the design would fail both conditions: a formatter handles diverse legitimate examples consistently and a missing value is represented without fabricating a string.
Question 240
The architect for a tender-response data extractor needs to ensure that an uncertain high-impact decision is routed for review instead of automatic execution. What should the team implement?
- reserve human handoff for specified triggers and let normal cases proceed
- proceed whenever the assistant's prose sounds certain
- mark the inconsistency and request resolution or human verification
- set a clear evidence or confidence boundary and require review below it
Explanation
Correct Answer
D:For a tender-response data extractor, risk-weighted gates limit autonomous actions when support is inadequate; this directly establishes that an uncertain high-impact decision is routed for review instead of automatic execution instead of relying on an unrelated workflow control.
Incorrect Answers
A:Although well-defined boundaries maintain autonomy without bypassing risky decisions, that only supports the distinct goal that an ordinary in-policy question is not escalated needlessly, not the requirement that an uncertain high-impact decision is routed for review instead of automatic execution in a tender-response data extractor.
B:In a tender-response data extractor, fluent language is not a calibrated reliability measure; as a result, the stated requirement that an uncertain high-impact decision is routed for review instead of automatic execution would still not be met.
C:Although conflicting source claims are a knowledge gap, not authorization to guess, that only supports the distinct goal that contradictory evidence is reviewed before action, not the requirement that an uncertain high-impact decision is routed for review instead of automatic execution in a tender-response data extractor.