Question 30
An incident postmortem for a competitive benchmarking pipeline requires controls so that the model can distinguish two similar business tools and that a new agent gets only durable context from a completed job. Which paired remedy covers both failure modes?
- give both tools vague names with identical generic descriptions; and initialize a fresh session with a verified structured summary
- give each tool a specific purpose and exclusion examples in its description; and initialize a fresh session with a verified structured summary
- give each tool a specific purpose and exclusion examples in its description; and paste a full historical transcript into every new session
- give both tools vague names with identical generic descriptions; and paste a full historical transcript into every new session
Explanation
Correct Answer
B:In a competitive benchmarking pipeline, discriminative descriptions reduce wrong routing when operations sound alike; alongside this, a bounded handoff transfers necessary facts without carrying discarded trial-and-error context, satisfying the distinct requirements that the model can distinguish two similar business tools and that a new agent gets only durable context from a completed job.
Incorrect Answers
A:In a competitive benchmarking pipeline, a bounded handoff transfers necessary facts without carrying discarded trial-and-error context, but a model cannot reliably select on boundaries that the interface never explains; therefore the separate requirement that the model can distinguish two similar business tools would remain unsatisfied.
C:In a competitive benchmarking pipeline, discriminative descriptions reduce wrong routing when operations sound alike, but irrelevant turns can consume context and revive outdated assumptions; therefore the separate requirement that a new agent gets only durable context from a completed job would remain unsatisfied.
D:In a competitive benchmarking pipeline, a model cannot reliably select on boundaries that the interface never explains and irrelevant turns can consume context and revive outdated assumptions; the design would fail both conditions: the model can distinguish two similar business tools and a new agent gets only durable context from a completed job.
Question 31
The architect for a warranty claims assistant needs to ensure that contradictory evidence is reviewed before action. What should the team implement?
- select the most recent-looking sentence without checking its source
- mark the inconsistency and request resolution or human verification
- ask a focused clarifying question for the missing requirement
- set a clear evidence or confidence boundary and require review below it
Explanation
Correct Answer
B:For a warranty claims assistant, conflicting source claims are a knowledge gap, not authorization to guess; this directly establishes that contradictory evidence is reviewed before action instead of relying on an unrelated workflow control.
Incorrect Answers
A:In a warranty claims assistant, recency alone cannot establish which contradictory record is correct; as a result, the stated requirement that contradictory evidence is reviewed before action would still not be met.
C:Although targeted clarification supplies the decision input before selecting an irreversible action, that only supports the distinct goal that a support case missing an essential fact is not silently guessed, not the requirement that contradictory evidence is reviewed before action in a warranty claims assistant.
D:Although risk-weighted gates limit autonomous actions when support is inadequate, that only supports the distinct goal that an uncertain high-impact decision is routed for review instead of automatic execution, not the requirement that contradictory evidence is reviewed before action in a warranty claims assistant.
Question 32
In a legacy service decomposition, the acceptance criteria require that a risky refactor is evaluated before any files change and that two nested packages follow distinct framework conventions. Which paired implementation meets both?
- let the agent immediately edit every matching file; and create separate path-specific rule files for each package
- implement only use plan mode to inspect dependencies and propose steps before execution; make no corresponding change for the second requirement
- use plan mode to inspect dependencies and propose steps before execution; and create separate path-specific rule files for each package
- use plan mode to inspect dependencies and propose steps before execution; and write one ambiguous instruction for both packages
- let the agent immediately edit every matching file; and write one ambiguous instruction for both packages
Explanation
Correct Answer
C:In a legacy service decomposition, planning separates investigation and approval from code mutation; alongside this, scoped conventions avoid contradictory patterns between codebases sharing a monorepo, satisfying the distinct requirements that a risky refactor is evaluated before any files change and that two nested packages follow distinct framework conventions.
Incorrect Answers
A:In a legacy service decomposition, scoped conventions avoid contradictory patterns between codebases sharing a monorepo, but early writes can spread an unapproved assumption through the repository; therefore the separate requirement that a risky refactor is evaluated before any files change would remain unsatisfied.
B:In a legacy service decomposition, planning separates investigation and approval from code mutation, but the independent mechanism for the other criterion is absent: scoped conventions avoid contradictory patterns between codebases sharing a monorepo; as written, the design does not ensure that two nested packages follow distinct framework conventions.
D:In a legacy service decomposition, planning separates investigation and approval from code mutation, but the model must guess which framework convention applies; therefore the separate requirement that two nested packages follow distinct framework conventions would remain unsatisfied.
E:In a legacy service decomposition, early writes can spread an unapproved assumption through the repository and the model must guess which framework convention applies; the design would fail both conditions: a risky refactor is evaluated before any files change and two nested packages follow distinct framework conventions.
Question 33
The architecture of a tender-response data extractor must meet these three criteria: fast-changing information is checked against its publication date; extracted records are labeled only when required fields agree; schema-valid output is also checked against source evidence. Choose THREE measures, one for each requirement.
- reuse an old cached claim without its date
- validate semantic constraints and provenance in application logic
- assume a schema-conforming response is factually correct
- classify as a match when any one keyword overlaps
- record source timestamps and refresh time-sensitive evidence
- state the conjunction of required matching conditions
Explanation
Correct Answers
B:For a tender-response data extractor, JSON shape does not prove that an extracted claim is supported by the document, which directly establishes that schema-valid output is also checked against source evidence; the separate requirement that fast-changing information is checked against its publication date is handled by a different action.
E:For a tender-response data extractor, freshness determines whether an otherwise credible fact applies now, which directly establishes that fast-changing information is checked against its publication date; the separate requirement that extracted records are labeled only when required fields agree is handled by a different action.
F:For a tender-response data extractor, explicit multi-field criteria prevent a single suggestive field from dominating the outcome, which directly establishes that extracted records are labeled only when required fields agree; the separate requirement that schema-valid output is also checked against source evidence is handled by a different action.
Incorrect Answers
A:For a tender-response data extractor, stale evidence may no longer describe the current state, so this option cannot establish that fast-changing information is checked against its publication date; correctly addressing extracted records are labeled only when required fields agree elsewhere does not repair that deficiency.
C:For a tender-response data extractor, syntactic conformance cannot guarantee source-level truth, so this option cannot establish that schema-valid output is also checked against source evidence; correctly addressing fast-changing information is checked against its publication date elsewhere does not repair that deficiency.
D:For a tender-response data extractor, a shared keyword does not prove record identity, so this option cannot establish that extracted records are labeled only when required fields agree; correctly addressing schema-valid output is also checked against source evidence elsewhere does not repair that deficiency.
Question 34
The release checklist for a repository maintenance copilot explicitly checks that dependent analysis begins only when source retrieval is complete; it also checks that untrusted project MCP definitions receive approval in interactive Claude Code sessions. What coordinated solution meets the checklist?
- run retrieval before analysis and pass its validated artifacts; and automatically approve every MCP definition found in a cloned repository
- start the analysis and retrieval agents together without shared results; and automatically approve every MCP definition found in a cloned repository
- run retrieval before analysis and pass its validated artifacts; and review and approve project-scoped server configuration before connecting interactively
- start the analysis and retrieval agents together without shared results; and review and approve project-scoped server configuration before connecting interactively
Explanation
Correct Answer
C:In a repository maintenance copilot, a downstream agent needs actual upstream findings rather than an assumed future result; alongside this, interactive project trust checks limit execution of unfamiliar server commands and endpoints, satisfying the distinct requirements that dependent analysis begins only when source retrieval is complete and that untrusted project MCP definitions receive approval in interactive Claude Code sessions.
Incorrect Answers
A:In a repository maintenance copilot, a downstream agent needs actual upstream findings rather than an assumed future result, but repository content can include unexpected executable commands and unattended sessions require explicit policy controls; therefore the separate requirement that untrusted project MCP definitions receive approval in interactive Claude Code sessions would remain unsatisfied.
B:In a repository maintenance copilot, the analysis branch cannot evaluate evidence it has not received and repository content can include unexpected executable commands and unattended sessions require explicit policy controls; the design would fail both conditions: dependent analysis begins only when source retrieval is complete and untrusted project MCP definitions receive approval in interactive Claude Code sessions.
D:In a repository maintenance copilot, interactive project trust checks limit execution of unfamiliar server commands and endpoints, but the analysis branch cannot evaluate evidence it has not received; therefore the separate requirement that dependent analysis begins only when source retrieval is complete would remain unsatisfied.
Question 35
While preparing a mobile client bug-fix task for deployment, engineers must ensure that a long workflow preserves confirmed facts separately from speculation; they must also ensure that secrets are not embedded in a shared MCP setup. Which combination best satisfies those requirements?
- mark verified, user-reported, and unknown state explicitly; and place long-lived access tokens directly in .mcp.json
- combine guesses and verified attributes in one unqualified summary; and place long-lived access tokens directly in .mcp.json
- combine guesses and verified attributes in one unqualified summary; and inject credentials through approved environment or secret-management mechanisms
- mark verified, user-reported, and unknown state explicitly; and inject credentials through approved environment or secret-management mechanisms
Explanation
Correct Answer
D:In a mobile client bug-fix task, epistemic labels stop a tentative suggestion becoming a confirmed action parameter; alongside this, separating credentials from versioned configuration limits accidental disclosure, satisfying the distinct requirements that a long workflow preserves confirmed facts separately from speculation and that secrets are not embedded in a shared MCP setup.
Incorrect Answers
A:In a mobile client bug-fix task, epistemic labels stop a tentative suggestion becoming a confirmed action parameter, but repository history can expose tokens to everyone with code access; therefore the separate requirement that secrets are not embedded in a shared MCP setup would remain unsatisfied.
B:In a mobile client bug-fix task, downstream operations may rely on an assumption as though it were checked and repository history can expose tokens to everyone with code access; the design would fail both conditions: a long workflow preserves confirmed facts separately from speculation and secrets are not embedded in a shared MCP setup.
C:In a mobile client bug-fix task, separating credentials from versioned configuration limits accidental disclosure, but downstream operations may rely on an assumption as though it were checked; therefore the separate requirement that a long workflow preserves confirmed facts separately from speculation would remain unsatisfied.
Question 36
An audit of a warranty claims assistant identifies two unmet requirements: refund execution is blocked until customer identity is verified and invalid parameters lead to a clear correction rather than a blind retry. Select TWO actions that remedy those particular deficiencies.
- hide the invalid field behind a generic internal-error message
- return a non-retryable validation error with the field and expected constraint
- enforce a prerequisite gate in application logic before the write tool
- ask the assistant in prose to remember verification first
Explanation
Correct Answers
B:For a warranty claims assistant, field-specific feedback lets the caller fix the input before resubmission, which directly establishes that invalid parameters lead to a clear correction rather than a blind retry; the separate requirement that refund execution is blocked until customer identity is verified is handled by a different action.
C:For a warranty claims assistant, deterministic preconditions protect an irreversible operation even if the model proposes it early, which directly establishes that refund execution is blocked until customer identity is verified; the separate requirement that invalid parameters lead to a clear correction rather than a blind retry is handled by a different action.
Incorrect Answers
A:For a warranty claims assistant, the client lacks the information needed to repair the request, so this option cannot establish that invalid parameters lead to a clear correction rather than a blind retry; correctly addressing refund execution is blocked until customer identity is verified elsewhere does not repair that deficiency.
D:For a warranty claims assistant, a prompt is advisory and does not technically block an unauthorized tool call, so this option cannot establish that refund execution is blocked until customer identity is verified; correctly addressing invalid parameters lead to a clear correction rather than a blind retry elsewhere does not repair that deficiency.
Question 37
An incident postmortem for a branch protection review bot requires controls so that a parser preserves the specified output structure across cases and that a developer has private preferences that should not change team guidance. Which paired remedy covers both failure modes?
- keep each demonstration's field order and types consistent; and add private preferences to the shared repository CLAUDE.md
- keep each demonstration's field order and types consistent; and place personal instructions in the user-level CLAUDE.md
- mix contradictory schemas in the example set; and place personal instructions in the user-level CLAUDE.md
- mix contradictory schemas in the example set; and add private preferences to the shared repository CLAUDE.md
Explanation
Correct Answer
B:In a branch protection review bot, coherent exemplars teach a stable mapping rather than competing formats; alongside this, global personal guidance is available to that user's sessions without committing preferences to the project, satisfying the distinct requirements that a parser preserves the specified output structure across cases and that a developer has private preferences that should not change team guidance.
Incorrect Answers
A:In a branch protection review bot, coherent exemplars teach a stable mapping rather than competing formats, but project instructions would impose one person's settings on the whole team; therefore the separate requirement that a developer has private preferences that should not change team guidance would remain unsatisfied.
C:In a branch protection review bot, global personal guidance is available to that user's sessions without committing preferences to the project, but inconsistent samples can make expected output ambiguous; therefore the separate requirement that a parser preserves the specified output structure across cases would remain unsatisfied.
D:In a branch protection review bot, inconsistent samples can make expected output ambiguous and project instructions would impose one person's settings on the whole team; the design would fail both conditions: a parser preserves the specified output structure across cases and a developer has private preferences that should not change team guidance.
Question 38
Before expanding a contract-analysis research network, the owner must demonstrate that a parser preserves the specified output structure across cases, while also demonstrating that a risk prompt handles missing supporting evidence honestly. Which design provides both controls?
- mix contradictory schemas in the example set; and require a definitive yes-or-no answer for every incomplete record
- keep each demonstration's field order and types consistent; and require a definitive yes-or-no answer for every incomplete record
- mix contradictory schemas in the example set; and instruct the model to return unknown or needs-review when evidence is absent
- keep each demonstration's field order and types consistent; and instruct the model to return unknown or needs-review when evidence is absent
Explanation
Correct Answer
D:In a contract-analysis research network, coherent exemplars teach a stable mapping rather than competing formats; alongside this, a defined abstention path avoids forced confident judgments without facts, satisfying the distinct requirements that a parser preserves the specified output structure across cases and that a risk prompt handles missing supporting evidence honestly.
Incorrect Answers
A:In a contract-analysis research network, inconsistent samples can make expected output ambiguous and forced binaries hide uncertainty as false certainty; the design would fail both conditions: a parser preserves the specified output structure across cases and a risk prompt handles missing supporting evidence honestly.
B:In a contract-analysis research network, coherent exemplars teach a stable mapping rather than competing formats, but forced binaries hide uncertainty as false certainty; therefore the separate requirement that a risk prompt handles missing supporting evidence honestly would remain unsatisfied.
C:In a contract-analysis research network, a defined abstention path avoids forced confident judgments without facts, but inconsistent samples can make expected output ambiguous; therefore the separate requirement that a parser preserves the specified output structure across cases would remain unsatisfied.
Question 39
Before expanding a multi-source policy brief service, the owner must demonstrate that a complex document receives focused quality checks, while also demonstrating that a broad tool catalog does not overwhelm a small task. Which design provides both controls?
- use one generic request to make the document better; and load all available MCP tools into every specialist turn
- separate factual, structural, and policy review passes; and load all available MCP tools into every specialist turn
- separate factual, structural, and policy review passes; and supply only context-relevant tools for the current agent
- use one generic request to make the document better; and supply only context-relevant tools for the current agent
Explanation
Correct Answer
C:In a multi-source policy brief service, specialized criteria can detect different failure modes without one overloaded evaluation instruction; alongside this, reducing irrelevant tool definitions improves selection and saves context, satisfying the distinct requirements that a complex document receives focused quality checks and that a broad tool catalog does not overwhelm a small task.
Incorrect Answers
A:In a multi-source policy brief service, an undifferentiated review can overlook specific required constraints and tool bloat makes the available actions harder to discriminate; the design would fail both conditions: a complex document receives focused quality checks and a broad tool catalog does not overwhelm a small task.
B:In a multi-source policy brief service, specialized criteria can detect different failure modes without one overloaded evaluation instruction, 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.
D:In a multi-source policy brief service, reducing irrelevant tool definitions improves selection and saves context, but an undifferentiated review can overlook specific required constraints; therefore the separate requirement that a complex document receives focused quality checks would remain unsatisfied.
Question 40
An incident postmortem for a dependency upgrade initiative requires controls so that a straightforward lookup avoids unnecessary agent coordination and that all contributors receive the repository's stable engineering conventions. Which paired remedy covers both failure modes?
- spawn a coordinator plus multiple specialists for every trivial query; and put concise shared guidance in the repository root CLAUDE.md
- use one focused tool-assisted agent for the bounded task; and put team-wide rules only in one developer's user-level instructions
- spawn a coordinator plus multiple specialists for every trivial query; and put team-wide rules only in one developer's user-level instructions
- use one focused tool-assisted agent for the bounded task; and put concise shared guidance in the repository root CLAUDE.md
Explanation
Correct Answer
D:In a dependency upgrade initiative, orchestration overhead is not justified when there is no useful specialization or concurrency; alongside this, project instructions travel with the codebase and establish common architecture and build constraints, satisfying the distinct requirements that a straightforward lookup avoids unnecessary agent coordination and that all contributors receive the repository's stable engineering conventions.
Incorrect Answers
A:In a dependency upgrade initiative, project instructions travel with the codebase and establish common architecture and build constraints, but extra agents add latency and handoff risk without new capabilities; therefore the separate requirement that a straightforward lookup avoids unnecessary agent coordination would remain unsatisfied.
B:In a dependency upgrade initiative, orchestration overhead is not justified when there is no useful specialization or concurrency, but personal files do not distribute the intended conventions to other contributors; therefore the separate requirement that all contributors receive the repository's stable engineering conventions would remain unsatisfied.
C:In a dependency upgrade initiative, extra agents add latency and handoff risk without new capabilities and personal files do not distribute the intended conventions to other contributors; the design would fail both conditions: a straightforward lookup avoids unnecessary agent coordination and all contributors receive the repository's stable engineering conventions.