Question 310 A compliance report requires every regulation to be mapped to controls. What decomposition makes completeness testable? Create a regulation-to-control matrix, verify each row, then synthesize gaps. Have each worker choose regulations without a shared index. Split solely by equal word counts. Write an executive summary first and declare it complete. Explanation Correct Answer A:Explicit coverage units make omissions visible before prose drafting. Incorrect Answers B:Uncoordinated choices can duplicate some and miss others. C:Text length does not correspond to legal control coverage. D:A polished summary cannot prove all regulations were mapped.
Question 311 An agent produces change-review summaries that mix confirmed bugs and speculative concerns. What would best improve calibration? Assume any code comment mentioning bug proves a runtime failure. Separate verified findings from hypotheses, stating evidence and validation needs for each. Present all hypotheses as confirmed defects. Delete uncertainty labels and require one confidence word only. Explanation Correct Answer B:Readers can act on confirmed defects and test uncertain ones. Incorrect Answers A:Comment text is not execution evidence. C:That misstates evidence strength. D:A bare label does not show the basis of a claim.
Question 312 A coding assistant needs to inspect one known config file and must not modify it. What is the most direct built-in tool? Write to overwrite the file with its expected contents. Bash to remove the file and recreate it. Read with the known path. Edit to replace an arbitrary line. Explanation Correct Answer C:Read retrieves file contents without editing the artifact. Incorrect Answers A:Overwriting destroys the evidence being inspected. B:That is unnecessarily destructive. D:The task is inspection, not mutation.
Question 313 A code assistant edits files in an unfamiliar repository and ignores the team’s “never modify generated files” rule. What should the team improve? Rename every generated file to a random filename. Rely on a developer remembering to mention the rule every turn. Disable all version control so changes cannot be committed. Put the generated-file boundary into relevant project guidance and enforce it via tooling where critical. Explanation Correct Answer D:A documented exclusion plus execution controls reduces unintended changes. Incorrect Answers A:Obscuring files does not communicate a safe editing policy. B:Repeated manual reminders are fragile. C:Removing history is not a protection against generated-file edits.
Question 314 A report-writing agent is given a strict two-minute deadline and can still improve its draft after that point. Which termination policy is appropriate? Continue indefinitely until the model declares perfection. Discard completed sections and deliver nothing whenever time expires. Use elapsed time alone as proof the draft is finished. Use normal protocol completion when possible, with deadline handling that marks partial work clearly. Explanation Correct Answer D:A deadline is a safety and service boundary; it must not be confused with substantive completion. Incorrect Answers A:Unbounded runs violate the service deadline. B:Preserving a clearly labeled partial draft is often more useful than silent data loss. C:Time expiry does not establish the report meets its requirements.
Question 315 An engineer adds citations to a strict JSON response mode that does not support interleaved citation blocks. What should the design change? Treat parser errors as successful extraction. Represent source references as schema fields or use a compatible non-JSON citation workflow. Ignore the API incompatibility and expect malformed mixed output to parse. Remove source provenance entirely. Explanation Correct Answer B:The response format must respect feature compatibility while retaining provenance. Incorrect Answers A:A failed response cannot be a valid record. C:Conflicting formats can trigger errors. D:The application still needs to support factual claims.
Question 316 An agent cannot reach a referenced primary document and only sees a search snippet. How should its statement be framed? Treat a search result title as proof of all related conclusions. Claim the full document has been reviewed. Quote imaginary paragraphs from the inaccessible document. Describe only what the snippet supports and disclose that the primary document was not inspected. Explanation Correct Answer D:Evidence limitations must not be concealed as source-level verification. Incorrect Answers A:Titles do not establish full content. B:That misrepresents access. C:That fabricates source text.
Question 317 Claude Code proposes three possible fixes but none has been tested. How should the team choose efficiently? Select one hypothesis using evidence, run focused validation, then refine based on results. Apply all three simultaneously without tracking changes. Choose whichever answer sounds most confident. Mark all proposals successful before implementation. Explanation Correct Answer A:Controlled experimentation distinguishes plausible fixes from working ones. Incorrect Answers B:Combined edits make causality and rollback difficult. C:Model confidence is not execution evidence. D:A suggestion is not a verified result.
Question 318 A policy research system launches one specialist for each country, then sends entire document dumps to the manager. Its manager context overflows. What contract should workers use? Return only a yes/no confidence label for every country. Return concise findings with citations and explicit open questions, keeping full documents referenced externally. Send raw documents twice so the coordinator sees all pages. Strip citations and keep untraceable summary prose. Explanation Correct Answer B:Structured summaries preserve decision-relevant evidence without flooding the coordinator. Incorrect Answers A:A label lacks the claims and evidence necessary for integration. C:Doubling raw context worsens the overflow. D:Missing provenance prevents validation and reliable synthesis.
Question 319 A pull-request reviewer calls every changed import a security vulnerability. What prompt instruction most directly increases finding precision? Classify every stylistic concern as critical severity. Require a concrete exploit path, affected resource, and code-line evidence for each security finding. Ask for at least 100 security comments per PR. Avoid looking at the changed code to remain unbiased. Explanation Correct Answer B:Evidence and impact criteria filter harmless syntactic changes from actionable vulnerabilities. Incorrect Answers A:Style alone is not proof of a security exposure. C:Volume targets incentivize false positives. D:Without the diff the reviewer lacks evidence for specific claims.
Question 320 A report generator tool offers summarize_document and extract_table. The assistant chooses summaries when it needs row-level values. What belongs in descriptions? An instruction that summaries always contain exact table cells. The same description for both because they read documents. A claim that all tools automatically validate user requirements. Distinct output contracts that explain summaries versus structured row extraction. Explanation Correct Answer D:The expected result type is decisive when selecting a tool. Incorrect Answers A:A summary need not retain every value. B:Shared domain does not mean equivalent outputs. C:The interface should not promise unimplemented guarantees.
