Question 20
For a mobile client bug-fix task, a reviewer identifies two independent obligations: separate code modules can be inspected in parallel and tool-call logging does not modify a successful tool result. Which proposal should be approved?
- implement only spawn independent specialists in the same delegation stage; make no corresponding change for the second requirement
- spawn independent specialists in the same delegation stage; and rewrite every tool result as a free-form paraphrase
- make the second specialist wait for the first even though their modules are independent; and rewrite every tool result as a free-form paraphrase
- make the second specialist wait for the first even though their modules are independent; and use an observation-only hook for telemetry when rewriting is unnecessary
- spawn independent specialists in the same delegation stage; and use an observation-only hook for telemetry when rewriting is unnecessary
Explanation
Correct Answer
E:In a mobile client bug-fix task, concurrent subagents can examine non-overlapping modules before the coordinator merges their reports; alongside this, separating instrumentation from mutation reduces the risk of corrupting evidence, satisfying the distinct requirements that separate code modules can be inspected in parallel and that tool-call logging does not modify a successful tool result.
Incorrect Answers
A:In a mobile client bug-fix task, concurrent subagents can examine non-overlapping modules before the coordinator merges their reports, but the independent mechanism for the other criterion is absent: separating instrumentation from mutation reduces the risk of corrupting evidence; as written, the design does not ensure that tool-call logging does not modify a successful tool result.
B:In a mobile client bug-fix task, concurrent subagents can examine non-overlapping modules before the coordinator merges their reports, but summaries can discard fields that subsequent steps require; therefore the separate requirement that tool-call logging does not modify a successful tool result would remain unsatisfied.
C:In a mobile client bug-fix task, serialization wastes time without supplying a dependency and summaries can discard fields that subsequent steps require; the design would fail both conditions: separate code modules can be inspected in parallel and tool-call logging does not modify a successful tool result.
D:In a mobile client bug-fix task, separating instrumentation from mutation reduces the risk of corrupting evidence, but serialization wastes time without supplying a dependency; therefore the separate requirement that separate code modules can be inspected in parallel would remain unsatisfied.
Question 21
An architecture review of a pull-request review pipeline finds two missing guarantees: only a nested service uses specialized implementation rules; a common security rule remains active across the project. What two-part change closes both gaps?
- place a dedicated CLAUDE.md in the relevant directory; and keep project-wide security guidance in shared instructions rather than a narrow path glob
- put all service-only details in the repository root instruction file; and scope the security instruction only to one unrelated test directory
- put all service-only details in the repository root instruction file; and keep project-wide security guidance in shared instructions rather than a narrow path glob
- place a dedicated CLAUDE.md in the relevant directory; and scope the security instruction only to one unrelated test directory
- implement only place a dedicated CLAUDE.md in the relevant directory; make no corresponding change for the second requirement
Explanation
Correct Answer
A:In a pull-request review pipeline, directory-specific context can refine instructions for work under that subtree; alongside this, cross-cutting obligations should not disappear when an agent changes directories, satisfying the distinct requirements that only a nested service uses specialized implementation rules and that a common security rule remains active across the project.
Incorrect Answers
B:In a pull-request review pipeline, globalizing narrow rules adds irrelevant instructions to unrelated work and other code paths would lack the universal control; the design would fail both conditions: only a nested service uses specialized implementation rules and a common security rule remains active across the project.
C:In a pull-request review pipeline, cross-cutting obligations should not disappear when an agent changes directories, but globalizing narrow rules adds irrelevant instructions to unrelated work; therefore the separate requirement that only a nested service uses specialized implementation rules would remain unsatisfied.
D:In a pull-request review pipeline, directory-specific context can refine instructions for work under that subtree, but other code paths would lack the universal control; therefore the separate requirement that a common security rule remains active across the project would remain unsatisfied.
E:In a pull-request review pipeline, directory-specific context can refine instructions for work under that subtree, but the independent mechanism for the other criterion is absent: cross-cutting obligations should not disappear when an agent changes directories; as written, the design does not ensure that a common security rule remains active across the project.
Question 22
An audit of a code-reference retrieval assistant asks how to make sure that near-overlapping retrieval tools are selected by actual boundary. Which implementation should be selected?
- declare an operation enum in the schema and reject values outside it in the tool handler
- document examples and conditions where each tool must not be used
- rely exclusively on short tool names for routing
- give each tool a specific purpose and exclusion examples in its description
Explanation
Correct Answer
B:For a code-reference retrieval assistant, contrastive guidance helps tool choice when resource names look similar; this directly establishes that near-overlapping retrieval tools are selected by actual boundary instead of relying on an unrelated workflow control.
Incorrect Answers
A:Although schema constraints guide valid selections and runtime validation blocks unsupported operation values, that only supports the distinct goal that a selector accepts only supported operation names, not the requirement that near-overlapping retrieval tools are selected by actual boundary in a code-reference retrieval assistant.
C:In a code-reference retrieval assistant, names alone may not capture the distinction in scope or authorization; as a result, the stated requirement that near-overlapping retrieval tools are selected by actual boundary would still not be met.
D:Although discriminative descriptions reduce wrong routing when operations sound alike, that only supports the distinct goal that the model can distinguish two similar business tools, not the requirement that near-overlapping retrieval tools are selected by actual boundary in a code-reference retrieval assistant.
Question 23
Before expanding a procurement research assistant, the owner must demonstrate that a conversational task does not invoke tools needlessly, while also demonstrating that an MCP tool failure can be traced across systems. Which design provides both controls?
- force a tool on every message regardless of intent; and replace every failure with the same unstructured unknown error
- use automatic tool choice when a call is optional; and propagate a stable correlation ID and machine-readable error code
- force a tool on every message regardless of intent; and propagate a stable correlation ID and machine-readable error code
- use automatic tool choice when a call is optional; and replace every failure with the same unstructured unknown error
Explanation
Correct Answer
B:In a procurement research assistant, auto allows the model to answer directly when no external side effect or fresh data is needed; alongside this, cross-service identifiers connect the user's symptom to upstream logs without exposing secrets, satisfying the distinct requirements that a conversational task does not invoke tools needlessly and that an MCP tool failure can be traced across systems.
Incorrect Answers
A:In a procurement research assistant, mandatory calls add latency and can invoke irrelevant operations and undifferentiated errors prevent meaningful triage; the design would fail both conditions: a conversational task does not invoke tools needlessly and an MCP tool failure can be traced across systems.
C:In a procurement research assistant, cross-service identifiers connect the user's symptom to upstream logs without exposing secrets, but mandatory calls add latency and can invoke irrelevant operations; therefore the separate requirement that a conversational task does not invoke tools needlessly would remain unsatisfied.
D:In a procurement research assistant, auto allows the model to answer directly when no external side effect or fresh data is needed, but undifferentiated errors prevent meaningful triage; therefore the separate requirement that an MCP tool failure can be traced across systems would remain unsatisfied.
Question 24
In a service-credit approval assistant, the design must ensure that a truncated response is not mistaken for successful completion and that a forced tool call still honors business authorization. Choose TWO independent changes that directly meet these two requirements.
- combine tool_choice with server-side policy checks
- interpret every non-tool stop reason as a successful final answer
- treat a forced tool choice as equivalent to user approval
- handle max_tokens as truncation rather than end_turn
Explanation
Correct Answers
A:For a service-credit approval assistant, selection controls when a tool is proposed but does not grant permission to execute, which directly establishes that a forced tool call still honors business authorization; the separate requirement that a truncated response is not mistaken for successful completion is handled by a different action.
D:For a service-credit approval assistant, reaching an output limit is a stop condition but does not establish task completion, which directly establishes that a truncated response is not mistaken for successful completion; the separate requirement that a forced tool call still honors business authorization is handled by a different action.
Incorrect Answers
B:For a service-credit approval assistant, a length-limited result may be incomplete or unusable, so this option cannot establish that a truncated response is not mistaken for successful completion; correctly addressing a forced tool call still honors business authorization elsewhere does not repair that deficiency.
C:For a service-credit approval assistant, tool selection and authorization are different controls, so this option cannot establish that a forced tool call still honors business authorization; correctly addressing a truncated response is not mistaken for successful completion elsewhere does not repair that deficiency.
Question 25
The release checklist for a source-grounded policy synthesis service explicitly checks that confidence scores reflect available evidence and uncertainty; it also checks that one agent's draft is reviewed independently for errors. What coordinated solution meets the checklist?
- calibrate confidence against known outcomes and source support; and ask the draft generator to declare its own work perfect
- calibrate confidence against known outcomes and source support; and use a separate review pass with explicit criteria
- set confidence to high whenever the explanation is detailed; and ask the draft generator to declare its own work perfect
- set confidence to high whenever the explanation is detailed; and use a separate review pass with explicit criteria
Explanation
Correct Answer
B:In a source-grounded policy synthesis service, observed error rates and evidence quality are better than a fluent self-rating; alongside this, independent criticism exposes omissions that unstructured self-praise can miss, satisfying the distinct requirements that confidence scores reflect available evidence and uncertainty and that one agent's draft is reviewed independently for errors.
Incorrect Answers
A:In a source-grounded policy synthesis service, observed error rates and evidence quality are better than a fluent self-rating, but self-assessment without a second criterion-based pass is weak evidence; therefore the separate requirement that one agent's draft is reviewed independently for errors would remain unsatisfied.
C:In a source-grounded policy synthesis service, verbal detail is not a validated reliability estimate and self-assessment without a second criterion-based pass is weak evidence; the design would fail both conditions: confidence scores reflect available evidence and uncertainty and one agent's draft is reviewed independently for errors.
D:In a source-grounded policy synthesis service, independent criticism exposes omissions that unstructured self-praise can miss, but verbal detail is not a validated reliability estimate; therefore the separate requirement that confidence scores reflect available evidence and uncertainty would remain unsatisfied.
Question 26
Before expanding a release-triage developer agent, the owner must demonstrate that a large research job combines specialized findings into one answer, while also demonstrating that a broad tool catalog does not overwhelm a small task. Which design provides both controls?
- use a coordinator that delegates focused tasks and owns final synthesis; and supply only context-relevant tools for the current agent
- use a coordinator that delegates focused tasks and owns final synthesis; and load all available MCP tools into every specialist turn
- have all subagents independently publish the final answer; and load all available MCP tools into every specialist turn
- have all subagents independently publish the final answer; and supply only context-relevant tools for the current agent
Explanation
Correct Answer
A:In a release-triage developer agent, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter; alongside this, reducing irrelevant tool definitions improves selection and saves context, satisfying the distinct requirements that a large research job combines specialized findings into one answer and that a broad tool catalog does not overwhelm a small task.
Incorrect Answers
B:In a release-triage developer agent, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter, 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.
C:In a release-triage developer agent, independent final outputs offer no single point to reconcile contradictions and tool bloat makes the available actions harder to discriminate; the design would fail both conditions: a large research job combines specialized findings into one answer and a broad tool catalog does not overwhelm a small task.
D:In a release-triage developer agent, reducing irrelevant tool definitions improves selection and saves context, but independent final outputs offer no single point to reconcile contradictions; therefore the separate requirement that a large research job combines specialized findings into one answer would remain unsatisfied.
Question 27
During review of a code policy conformance job, an unmet criterion states that policy decisions use measurable conditions rather than tone. Which measure resolves this specific gap?
- define explicit positive evidence and an abstain or review category
- state the conjunction of required matching conditions
- describe exact observable triggers and exception rules
- ask the model to use its general intuition about severity
Explanation
Correct Answer
C:For a code policy conformance job, operationalized criteria reduce disagreement across similar inputs; this directly establishes that policy decisions use measurable conditions rather than tone instead of relying on an unrelated workflow control.
Incorrect Answers
A:Although clear decision thresholds separate direct evidence from weak similarity, that only supports the distinct goal that a classifier avoids marking uncertain cases as confirmed violations, not the requirement that policy decisions use measurable conditions rather than tone in a code policy conformance job.
B:Although explicit multi-field criteria prevent a single suggestive field from dominating the outcome, that only supports the distinct goal that extracted records are labeled only when required fields agree, not the requirement that policy decisions use measurable conditions rather than tone in a code policy conformance job.
D:In a code policy conformance job, unquantified intuition varies between cases and reviewers; as a result, the stated requirement that policy decisions use measurable conditions rather than tone would still not be met.
Question 28
In a technical due-diligence report generator, the acceptance criteria require that conflicting subagent findings produce one accountable interpretation and that the report is not drafted before evidence is checked. Which paired implementation meets both?
- let the fastest subagent override other findings automatically; and start drafting at the same time as the source search and assume success
- have the coordinator compare claims and sources before synthesis; and place evidence validation before the drafting stage
- have the coordinator compare claims and sources before synthesis; and start drafting at the same time as the source search and assume success
- let the fastest subagent override other findings automatically; and place evidence validation before the drafting stage
Explanation
Correct Answer
B:In a technical due-diligence report generator, central reconciliation can distinguish evidence disagreement from a formatting difference; alongside this, a dependent stage must receive validated inputs to avoid propagating unsupported conclusions, satisfying the distinct requirements that conflicting subagent findings produce one accountable interpretation and that the report is not drafted before evidence is checked.
Incorrect Answers
A:In a technical due-diligence report generator, completion order is not a sound evidence-ranking rule and the writer may treat unverified or absent evidence as real; the design would fail both conditions: conflicting subagent findings produce one accountable interpretation and the report is not drafted before evidence is checked.
C:In a technical due-diligence report generator, central reconciliation can distinguish evidence disagreement from a formatting difference, but the writer may treat unverified or absent evidence as real; therefore the separate requirement that the report is not drafted before evidence is checked would remain unsatisfied.
D:In a technical due-diligence report generator, a dependent stage must receive validated inputs to avoid propagating unsupported conclusions, but completion order is not a sound evidence-ranking rule; therefore the separate requirement that conflicting subagent findings produce one accountable interpretation would remain unsatisfied.
Question 29
A team operating a procurement document processing system wants to establish two separate behaviors: persistent ambiguous data receives human review; in addition, a risk prompt handles missing supporting evidence honestly. Which plan implements both?
- keep retrying until the model finally emits something that passes; and require a definitive yes-or-no answer for every incomplete record
- stop after defined retries and route unresolved cases to a reviewer; and instruct the model to return unknown or needs-review when evidence is absent
- keep retrying until the model finally emits something that passes; and instruct the model to return unknown or needs-review when evidence is absent
- stop after defined retries and route unresolved cases to a reviewer; and require a definitive yes-or-no answer for every incomplete record
Explanation
Correct Answer
B:In a procurement document processing system, a bounded escalation path prevents infinite loops and invented resolutions; alongside this, a defined abstention path avoids forced confident judgments without facts, satisfying the distinct requirements that persistent ambiguous data receives human review and that a risk prompt handles missing supporting evidence honestly.
Incorrect Answers
A:In a procurement document processing system, unlimited retries waste resources and can reinforce guesses and forced binaries hide uncertainty as false certainty; the design would fail both conditions: persistent ambiguous data receives human review and a risk prompt handles missing supporting evidence honestly.
C:In a procurement document processing system, a defined abstention path avoids forced confident judgments without facts, but unlimited retries waste resources and can reinforce guesses; therefore the separate requirement that persistent ambiguous data receives human review would remain unsatisfied.
D:In a procurement document processing system, a bounded escalation path prevents infinite loops and invented resolutions, but forced binaries hide uncertainty as false certainty; therefore the separate requirement that a risk prompt handles missing supporting evidence honestly would remain unsatisfied.
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.