Question 270
A multi-source intelligence assistant must preserve provenance in its conclusions. An OCR system extracts text from a scanned image with uncertain characters. What should the team do to resolve this particular gap?
- Record the direct document locator and relevant passage reference
- Label the date unknown and qualify conclusions that depend on recency
- Record file type, version, denominator, and calculation method for each source
- Store each value alongside publication and observation dates
- Mark the source as OCR-derived and retain the image or page locator
Explanation
Correct Answer
E:Image-derived text can contain recognition errors; noting origin supports appropriate human verification. For the stated case, the check is whether the system can preserve content modality and extraction status; this approach supplies that control.
Incorrect Answers
A:A generic homepage does not let reviewers trace the asserted evidence; precise provenance makes claims auditable; the alternative would make sense if a summary cites the homepage of a website but not the page containing a key claim. It does not directly fix the observed condition: an OCR system extracts text from a scanned image with uncertain characters.
B:When a retrieved document lacks a reliable publication date, this would be useful because a missing timestamp cannot be repaired by inference from a filename or crawl date; time-sensitive claims need verified dates; this item instead calls for a way to preserve content modality and extraction status.
C:This measure fits a different trigger: a CSV report and a narrative article disagree about a count. In that setting, A discrepancy may arise from different populations or transformations; provenance details are needed before choosing a value; it does not resolve the present need to preserve content modality and extraction status.
D:Different dates may describe genuine changes rather than factual contradiction; temporal provenance enables accurate interpretation; the relevant precondition would be two sources give different values for the same metric on different dates. The question instead describes an OCR system extracts text from a scanned image with uncertain characters.
Question 271
A CI pipeline runs Claude Code non-interactively for pull-request review. A pull request from an untrusted fork contains instructions to exfiltrate secrets. Which control is most directly relevant?
- Use an independent check or human approval before merging
- Run Claude Code in noninteractive print mode with a bounded task prompt
- Run in a restricted sandbox with minimal tokens and no secret-bearing privileged steps
- Set job timeout, token or iteration budgets, and explicit failure behavior
Explanation
Correct Answer
C:Pull-request content is untrusted input; prompt instructions cannot substitute for runner isolation or secret scoping.
Incorrect Answers
A:This measure fits a different trigger: the team wants Claude's review to recommend fixes but not approve its own changes. In that setting, A model reviewing its own output is correlated with its mistakes; a distinct gate reduces unchecked self-approval; it does not resolve the present need to protect CI credentials from model-controlled code.
B:A headless execution path avoids UI expectations and lets CI capture the completed response as an artifact; the relevant precondition would be the build runner has no interactive terminal for answering Claude prompts. The question instead describes a pull request from an untrusted fork contains instructions to exfiltrate secrets.
D:When a CI task might continue indefinitely under repeated tool calls, this would be useful because CI should fail visibly when budgets are exhausted rather than hang or silently return partial success; this item instead calls for a way to protect CI credentials from model-controlled code.
Question 272
A support assistant drafts responses that must meet measurable service standards. A support reply must acknowledge an issue and give the next permitted step. How should the workflow be adjusted?
- Require policy identifier, effective date, and relevant decision rationale
- Specify acknowledgement, verified status, and a concrete authorized next action
- Require explicit uncertainty and an escalation route when no sourced estimate exists
- Specify required facts, tone limits, forbidden claims, and measurable completion checks
Explanation
Correct Answer
B:Component-level criteria make the outcome testable while leaving room for natural wording within the approved service tone.
Incorrect Answers
A:When a regulator requires every decision note to cite the policy version used, this would be useful because mandatory citation fields connect conclusions to the governing rule and make later audits possible when policies change; this item instead calls for a way to define required response components.
C:This measure fits a different trigger: the assistant invents a delivery estimate when the carrier has not supplied one. In that setting, A safe prompt states what to do when evidence is absent; a request for confidence or completeness can otherwise encourage fabrication; it does not resolve the present need to define required response components.
D:Explicit criteria guide generation and review; subjective adjectives alone do not define whether a reply satisfied the support policy; the relevant precondition would be reviewers disagree because the instruction merely says 'be helpful.'. The question instead describes a support reply must acknowledge an issue and give the next permitted step.
Question 273
An AI research assistant prepares high-stakes organizational recommendations. A requested final verdict depends on a policy that has not been supplied. Which implementation decision is the best fit?
- Mark the affected conclusion as unverified and request further review
- Execute the delegated decision inside the defined thresholds with an audit trail
- Request the applicable policy or route to its owner
- Apply documented style rules automatically and continue
Explanation
Correct Answer
C:An agent cannot safely invent an organization's normative requirements from generic patterns. This matches the workflow goal to ask for the governing rule before deciding rather than introducing an unrelated action.
Incorrect Answers
A:The absence of a crucial source limits what the assistant can responsibly conclude; a plausible summary cannot replace verification; the relevant precondition would be a key source is inaccessible and no equivalent evidence is available. The question instead describes a requested final verdict depends on a policy that has not been supplied.
B:When a legal owner has explicitly delegated a narrow approved judgment, this would be useful because escalation should follow actual authority boundaries; needless handoffs for approved routine cases slow the system without improving safety; this item instead calls for a way to ask for the governing rule before deciding.
D:This measure fits a different trigger: a minor stylistic issue does not affect factual or safety requirements. In that setting, Human review capacity is best reserved for consequential ambiguity; deterministic cosmetic corrections can be handled within the workflow; it does not resolve the present need to ask for the governing rule before deciding.
Question 274
A development team wants consistent Claude Code behavior across a shared monorepo. A single package has unusual testing requirements not relevant elsewhere. What should the architect specify in the design?
- Keep a short index of essential rules and reference detailed docs only when needed
- Place package-specific guidance near the relevant subproject or use path-scoped rules
- Keep the personal preference in user-level Claude Code guidance
- Place concise project instructions in a committed project CLAUDE.md file
Explanation
Correct Answer
B:Local rules reduce noise in unrelated tasks and clarify when a special test command actually applies. This matches the workflow goal to scope instructions to a subtree rather than introducing an unrelated action.
Incorrect Answers
A:This measure fits a different trigger: the main CLAUDE.md duplicates a lengthy internal engineering handbook. In that setting, Large generic guidance competes with working context; focused instructions are easier for Claude to use reliably; it does not resolve the present need to scope instructions to a subtree.
C:User-level configuration supports individual habits without imposing personal formatting conventions on the entire repository; the alternative would make sense if one developer prefers a personal response style that colleagues do not share. It does not directly fix the observed condition: a single package has unusual testing requirements not relevant elsewhere.
D:Project instructions make conventions available when working in the repository; they should describe relevant workflow rules rather than replicate all documentation; the relevant precondition would be every contributor must follow the same build and review commands for the repository. The question instead describes a single package has unusual testing requirements not relevant elsewhere.
Question 275
A coding assistant reviews generated patch descriptions for consistency. The assistant copies placeholder identifiers from example output. Which change most directly addresses this condition?
- Annotate samples with the behavior or contract difference that changes the classification
- Provide few-shot examples that distinguish behavior-preserving and behavior-changing patches
- Use clearly illustrative placeholders and require current identifiers from the input
- Add a small labeled example covering coordinated changes across multiple modules
- Revise labels and re-evaluate examples against the active review criteria
Explanation
Correct Answer
C:Few-shot examples guide format but must not become a source of fabricated ticket IDs or commit hashes. The relevant requirement is to separate examples from required live values, and this measure operates at that decision point.
Incorrect Answers
A:The important signal is semantic change, not file size or formatting; rationale-bearing examples make that visible; the alternative would make sense if a reviewer cannot tell why two visually similar changes receive different labels. It does not directly fix the observed condition: the assistant copies placeholder identifiers from example output.
B:Contrastive labels teach the category distinction that ambiguous prose alone may not reliably convey; the relevant precondition would be some changes are wrongly described as refactors rather than bug fixes. The question instead describes the assistant copies placeholder identifiers from example output.
D:This measure fits a different trigger: the team's examples omit multi-file changes. In that setting, A representative complex case teaches how to summarize a unit of work that spans files without treating each edit as unrelated; it does not resolve the present need to separate examples from required live values.
E:When an example is accurate for old review policy but not the current standard, this would be useful because examples are policy-bearing inputs; inconsistency between versions can undermine otherwise precise instructions; this item instead calls for a way to separate examples from required live values.
Question 276
An agent runs a bounded support workflow with multiple optional tool calls. A backend becomes rate limited while many agents share it. What is the most defensible next step?
- Return a bounded handoff or clarification rather than forcing more autonomous turns
- Enforce a maximum tool-call or iteration budget and a fallback outcome
- Use rate-aware concurrency limits and jittered backoff
- Log authorized tool names, request IDs, outcomes, and decision context
Explanation
Correct Answer
C:Application-side scheduling controls aggregate demand; independent immediate retries can amplify throttling across sessions.
Incorrect Answers
A:When a response contains no tool calls but the task is still unresolved and requires human judgment, this would be useful because when prerequisites are missing, another iteration does not create evidence; explicit unresolved state protects correctness and enables review; this item instead calls for a way to avoid synchronized retry storms.
B:Without a host-side bound, repeatedly proposing the same action can consume resources without resolving the customer's issue; the relevant precondition would be a tool-driven conversation keeps calling the same read-only endpoint without progress. The question instead describes a backend becomes rate limited while many agents share it.
D:This measure fits a different trigger: support needs a record of why an agent issued each sensitive call. In that setting, Structured execution logs establish which external action occurred and on what evidence without assuming that model prose is a reliable audit record; it does not resolve the present need to avoid synchronized retry storms.
Question 277
An organization uses Claude to summarize an archive of technical incidents. A run returns finished summaries but no mapping to original incident IDs. Which approach satisfies the stated constraint?
- Verify stored document digest and cited spans before accepting the result
- Process prerequisite records before dispatching dependent summaries
- Include stable incident IDs in each request and result mapping
- Use batch processing and an output manifest for the incident collection
Explanation
Correct Answer
C:Without correlation keys, a bulk response can be structurally complete but impossible to assign safely to the correct incident.
Incorrect Answers
A:This measure fits a different trigger: one summary appears to describe a different source document. In that setting, Matching output to the right input is a semantic and provenance requirement, not merely a batch-completion check; it does not resolve the present need to repair result correlation metadata.
B:Batching does not remove ordering requirements; data-dependent reports need validated inputs from earlier stages; the alternative would make sense if some incidents must be summarized in chronological dependency order. It does not directly fix the observed condition: a run returns finished summaries but no mapping to original incident IDs.
D:Large independent jobs can be processed without tying up a conversational request; the manifest preserves record identity; the relevant precondition would be the workload has many independent records and no user waiting for each response. The question instead describes a run returns finished summaries but no mapping to original incident IDs.
Question 278
A release team operates Claude Code as part of automated change analysis. A reviewer finds a high-risk issue requiring a person's decision. Which mechanism should be selected?
- Emit a fixed verdict schema and reject unparseable output
- Block automatic merge and provide evidence for maintainer review
- Compare findings against adjudicated examples and track precision and recall
- Re-run build and tests from the committed revision in a clean environment
Explanation
Correct Answer
B:A high-risk finding should not be auto-dismissed by the same generative process that produced the report. This matches the workflow goal to route risk into a human-controlled approval gate rather than introducing an unrelated action.
Incorrect Answers
A:A deterministic parser must be able to identify pass, needs-review, and blocked states without interpreting conversational language; the relevant precondition would be the pipeline prints a detailed narrative but downstream jobs need a verdict field. The question instead describes a reviewer finds a high-risk issue requiring a person's decision.
C:This measure fits a different trigger: the model-generated review itself must be measured for false positives over time. In that setting, CI model output can drift; longitudinal labeled evaluation measures whether the review remains useful rather than merely verbose; it does not resolve the present need to route risk into a human-controlled approval gate.
D:Verification from a clean checkout detects hidden local state and makes results reproducible across runners; the alternative would make sense if an agent proposes a code change that CI cannot reproduce in a fresh checkout. It does not directly fix the observed condition: a reviewer finds a high-risk issue requiring a person's decision.
Question 279
A monorepo uses different coding guidelines for backend and frontend files. A global style rule applies to every source file in the repository. What should be implemented at the appropriate boundary?
- Narrow the patterns or align their formatter requirements
- Scope the testing guidance to Python service paths
- Create a path-scoped rule targeting the web source patterns
- Keep the common style principle in root project instructions
Explanation
Correct Answer
D:Shared behavior should have one clear home rather than several overlapping path rules that can contradict one another.
Incorrect Answers
A:When two path rules overlap and recommend incompatible formatters, this would be useful because overlapping rule sets should lead to one predictable tool choice; unresolved conflicts create patch churn; this item instead calls for a way to avoid duplicating truly global guidance.
B:A service-specific rule prevents Claude from running a frontend command on an unrelated backend change; the alternative would make sense if python services require an alternate unit-test command. It does not directly fix the observed condition: a global style rule applies to every source file in the repository.
C:Path-specific guidance appears for relevant work instead of affecting backend files that use different tooling; the relevant precondition would be typeScript files under web/ require browser-specific linting rules. The question instead describes a global style rule applies to every source file in the repository.
Question 280
An internal compliance system tracks where evidence for decisions originated. An extraction result is copied into a report with no link to its document. Which option best meets the operational need?
- Record retrieval time and a content digest or archived excerpt where permitted
- Carry source ID, page, and extraction run identifier into the final record
- Compare source-of-truth ownership and freshness before selecting a status
- Mark quotations, extracted facts, and analyst inferences separately
- Attach the document version and effective date to the quoted passage
Explanation
Correct Answer
B:A provenance chain connects the published value to the observed document and the specific processing run that produced it.
Incorrect Answers
A:When a cited webpage changes after the original analysis was completed, this would be useful because web evidence is mutable; a stable reference explains what the system actually saw when reaching its original conclusion; this item instead calls for a way to retain lineage through transformations.
C:A newer timestamp is not always sufficient if one system is a derived cache; authority and synchronization matter; the alternative would make sense if two tools return different customer statuses from separate data stores. It does not directly fix the observed condition: an extraction result is copied into a report with no link to its document.
D:This measure fits a different trigger: a research report mixes direct quotations and model paraphrases without labeling them. In that setting, Separating evidence types prevents a model's synthesis from appearing to be a verbatim statement by an external authority; it does not resolve the present need to retain lineage through transformations.
E:Policy text changes over time; revision identity prevents a once-valid rule from being applied to the wrong period; the relevant precondition would be a policy interpretation quotes a paragraph but does not identify the policy revision. The question instead describes an extraction result is copied into a report with no link to its document.