Question 100
A human reviewer is inundated with thousands of low-risk extraction exceptions. What routing policy improves scalability?
- Mandate evidence and calibration rather than trusting the self-reported confidence number
- Present policy basis, supporting records, and the requested action before approval
- Hold back from automatic commitment and send the exception to the authorized owner
- Prioritize by material impact, uncertainty and business risk
- Keep individual rationales, evidence and the final adjudication basis
Explanation
Correct Answer
D:Triage uses the cost of errors to allocate scarce review capacity. Not all missing fields need the same urgency or reviewer expertise.
Incorrect Answers
A:A model's expressed confidence may not correspond to measured probability; that serves avoiding uncalibrated confidence as proof, whereas the case requires directing human attention to the highest-risk cases.
B:Because meaningful approval requires adequate context to judge eligibility and consequences, the measure supports giving human reviewers sufficient decision evidence rather than the needed directing human attention to the highest-risk cases.
C:Policy ambiguity is not resolved by fluent reasoning; its benefits for routing uncertain policy exceptions to authorized review do not resolve the need for directing human attention to the highest-risk cases.
E:Disagreement may expose ambiguous policies or missing facts; this contributes to retaining review disagreement and adjudication provenance without meeting the need for directing human attention to the highest-risk cases.
Question 101
An upstream service returns a diagnostic message that contains private authentication data. What should the parent expose to other agents?
- Propagate a sanitized error category and protected correlation reference
- Reject the publication gate whenever mandatory validation fails
- Differentiate tool failure from a valid search with zero matches
- Capture the evidence gap and either retry safely or abstain from the unsupported claim
Explanation
Correct Answer
A:Error propagation should preserve recovery information without spreading secrets. Full internal logs belong in restricted telemetry, not general agent contexts.
Incorrect Answers
B:Successful generation is distinct from successful verification; a method suited to honoring validation failures during workflow completion does not meet the need for redacting sensitive details in propagated errors.
C:For propagating search errors without treating them as no-match evidence, infrastructure errors must propagate as errors, not negative evidence; however, this requirement concerns redacting sensitive details in propagated errors.
D:Failed retrieval does not authorize fabricating authoritative state; that may serve preventing fabricated facts after upstream errors, but it is no substitute for redacting sensitive details in propagated errors.
Question 102
A pipeline attempts to stream individual model tokens while submitting a Message Batch. What design change is necessary?
- Keep outcomes by input ID and create a retry queue from non-success statuses
- Remove streaming from batch requests and process completed result artifacts
- Check pending request limits and pace submissions by available capacity
- Keep each request's custom_id and associated error classification
Explanation
Correct Answer
B:Batch execution yields collected request outcomes rather than an interactive token stream. A consumer should retrieve results using the batch workflow.
Incorrect Answers
A:Although structured reconciliation lets teams preserve successful results and focus intervention on failures, the choice addresses constructing targeted post-batch retries and not the requested using supported response delivery for batch processing.
C:Batch queues remain subject to provider limits and account constraints; the option would serve respecting throughput limits when batching, not using supported response delivery for batch processing as specified in the stem.
D:Stable item identifiers connect failures to inputs and retry history; this addresses linking each batch error to its source record, while the question is about using supported response delivery for batch processing.
Question 103
A reviewer wants to test an alternative plan without changing the original agent session. Which state operation is appropriate?
- Keep message order and essential tool-result linkage along with session metadata
- Save only necessary state and apply an explicit retention or deletion policy
- Verify the fork against the current codebase and rerun relevant checks
- Fork the existing session into an isolated alternative branch
Explanation
Correct Answer
D:Forking supports experimentation from a shared starting point while separating subsequent messages. Appending tests to the original transcript would alter its decision history.
Incorrect Answers
A:When considering checkpointing structured agent progress, resumable state includes structured execution history, not merely the final natural-language summary; the requirement for branching session state without overwriting the original remains unmet.
B:When addressing limiting persisted session data, resumability does not justify indefinite storage of sensitive context; the present task instead calls for branching session state without overwriting the original.
C:A successful result on an older branch does not prove current compatibility; validating branch changes before merge and branching session state without overwriting the original require different controls in this situation.
Question 104
A project must create a new generated README file whose destination path does not yet exist. Which built-in capability fits?
- Employ Glob to enumerate files matching the path pattern
- Execute the formatter through Bash and inspect the resulting diff or file changes
- Use Glob to find the matching migration paths and then inspect them in order
- Use Write to create the required file content
Explanation
Correct Answer
D:Write creates or replaces file contents at a chosen path. This is suitable for a complete new artifact, but requires care not to overwrite unrelated existing work.
Incorrect Answers
A:Glob resolves matching file paths by name or path shape; the action supports enumerating files by filename pattern but falls short of creating a new source-controlled document in this case.
B:Process output alone may not prove exactly what changed; this can serve verifying changes produced by an executable command, but leaves the requirement for creating a new source-controlled document unmet.
C:Filename enumeration establishes the migration set without executing anything; that addresses gathering migration files without executing code, whereas this case calls for creating a new source-controlled document.
Question 105
A generated change request includes required approvals for two business owners. What should the controller do before applying it?
- Add a human approval gate between proposal and execution
- Include an executable policy gate before the final approval or refund call
- Specify a post-action reconciliation stage before marking the case resolved
- Verify all required approvals as structured preconditions for execution
- Focus on making eligibility verification a required predecessor in the workflow graph
Explanation
Correct Answer
D:Multi-party authorization should be checked against the configured policy, not inferred from a narrative. Missing any mandatory approval must block the side effect.
Incorrect Answers
A:Human-in-the-loop approval must be enforced by the controller for side effects; its benefits for inserting human approval before side effects do not resolve the need for enforcing multi-party approval prerequisites.
B:A deterministic guard can compare amounts with configured limits; supporting enforcing financial limits in a workflow cannot replace the required enforcing multi-party approval prerequisites.
C:Workflow success should depend on confirmed downstream state, not just a tool invocation; this contributes to verifying all required stages of case resolution without meeting the need for enforcing multi-party approval prerequisites.
E:Because an explicit dependency ensures the mutation cannot begin before policy checks succeed, the measure supports enforcing refund prerequisites before mutation rather than the needed enforcing multi-party approval prerequisites.
Question 106
Two reviewers reach incompatible conclusions about a security finding. What should the reconciliation pass do?
- Compare cited evidence and apply an explicit resolution rubric
- Launch an independent verifier against the original requirements and observed tool evidence
- Prioritize findings by impact and evidence strength before producing recommendations
- Preserve validated facts and limit later passes to approved presentation edits
Explanation
Correct Answer
A:Disagreement is information to evaluate, not a reason to average opinions blindly. An evidence-based adjudicator can label unresolved cases for human review.
Incorrect Answers
B:Self-assessment tends to reuse the model's assumptions; that may serve reducing self-confirmation bias in generated artifacts, but it is no substitute for reconciling contradictory reviewer findings.
C:Review output should prioritize well-supported important issues; this may support prioritizing findings after independent review but cannot alone meet the requirement for reconciling contradictory reviewer findings.
D:Separating factual content from stylistic changes reduces accidental meaning drift; the stated problem requires reconciling contradictory reviewer findings rather than preserving verified facts during editing.
Question 107
A summarizer merges decisions made by different meeting attendees into one unqualified recommendation. What should the instruction require?
- Specify evaluation priorities such as latency, correctness, budget and operational ownership
- Attribute decisions and proposals to named sources, preserving unresolved disagreement
- Insist on source-backed claims and a labeled unknowns or uncertainty section
- Specify task scope, stopping conditions and prohibited side quests
- Formalize redaction rules and permitted data fields in the expected summary
Explanation
Correct Answer
B:Attribution is part of factual correctness. An explicit decision ledger prevents the output from treating debate as unanimous agreement.
Incorrect Answers
A:Architectural choices depend on goals and trade-offs; the option would serve making architecture trade-off priorities explicit, not preserving attribution in generated meeting summaries as specified in the stem.
C:Requiring evidence mapping and explicit uncertainty discourages treating unverified inference as fact; that relates to separating sourced conclusions from unknowns, not the necessary preserving attribution in generated meeting summaries.
D:Although bounded instructions tell the agent what enough evidence looks like, the choice addresses defining a stopping condition for research and not the requested preserving attribution in generated meeting summaries.
E:Data minimization instructions make the output contract sensitive to privacy; the deciding condition is preserving attribution in generated meeting summaries, whereas this action concerns minimizing sensitive data in summaries.
Question 108
A research assistant finds only one weak source for a high-stakes technical assertion. How should it write the conclusion?
- Qualify the claim and request stronger corroboration before treating it as established
- Mark confirmed observations separately from estimates and unknowns
- Document retrieval time and a source version or approved snapshot reference
- Demand a matching source span or flag the clause as unsupported
Explanation
Correct Answer
A:Weak or incomplete evidence warrants explicit uncertainty. Overstated certainty can make downstream decisions rely on a proposition that has not been verified.
Incorrect Answers
B:Provenance distinguishes system-of-record data from inference; the unsatisfied requirement is calibrating claims to evidence strength, not separating verified status from estimates.
C:Live pages can change or disappear; in this scenario, preserving time-sensitive research provenance is secondary to calibrating claims to evidence strength.
D:When considering tracing extracted clauses to document spans, evidence-grounded extraction ties fields to document locations; the requirement for calibrating claims to evidence strength remains unmet.
Question 109
A patch fixes behavior but introduces an uncovered error path with no fallback. What refinement is required?
- Choose failing integration output to refine the patch and rerun relevant tests
- Confirm behavioral acceptance cases rather than treating lint success as completion
- Exercise failure-path tests and add explicit handling where warranted
- Choose the smallest clear change that meets requirements and preserves interfaces
Explanation
Correct Answer
C:Reliable implementation must address realistic negative inputs and dependency errors. Happy-path success alone does not establish production readiness.
Incorrect Answers
A:Iterative development uses tool feedback as evidence; that addresses refining code based on integration-test feedback, whereas this case calls for validating negative paths after implementation.
B:Linting checks style and some static issues; it does not prove runtime correctness; this concerns distinguishing static hygiene from functional success, leaving the requirement for validating negative paths after implementation outstanding.
D:Simplicity is an engineering constraint: extra layers need a demonstrated benefit; the action supports reviewing unnecessary complexity in generated code but falls short of validating negative paths after implementation in this case.
Question 110
A workflow has hooks that log errors but does not stop a failed authorization check. What change is required?
- Specify ordered responsibilities and nonconflicting hook contracts
- Return a blocking control result instead of an informational log event
- Execute a PostToolUse handler after successful file-edit tools
- Attach the validation to the agent-turn or stop lifecycle event
- Enforce a PreToolUse hook that denies matching unsafe invocations
Explanation
Correct Answer
B:Emitting telemetry alone does not enforce permission decisions. The hook must use the runtime's supported deny mechanism at the correct stage.
Incorrect Answers
A:Clear handler scope and explicit failure precedence prevent accidental interactions; this contributes to resolving conflicting lifecycle handlers without meeting the need for distinguishing logging from access denial.
C:Because a post-tool callback can inspect the completed edit and trigger formatting or verification in response, the measure supports reacting to a completed editing tool rather than the needed distinguishing logging from access denial.
D:End-of-turn hooks can trigger aggregate checks that would be wasteful after every individual edit; its benefits for validating at the end of agent activity do not resolve the need for distinguishing logging from access denial.
E:PreToolUse runs before a tool call and can participate in the decision to block it; supporting blocking disallowed commands before tool execution cannot replace the required distinguishing logging from access denial.