Question 730
Production readiness for an order-cancellation service depends on ensuring that a new agent gets only durable context from a completed job and that partial completion of a multi-step operation is handled safely. Select the most appropriate combined approach.
- paste a full historical transcript into every new session; and treat the first successful tool call as proof that the entire workflow succeeded
- initialize a fresh session with a verified structured summary; and track individual stage outcomes and use explicit recovery or compensating actions
- paste a full historical transcript into every new session; and track individual stage outcomes and use explicit recovery or compensating actions
- initialize a fresh session with a verified structured summary; and treat the first successful tool call as proof that the entire workflow succeeded
Explanation
Correct Answer
B:In an order-cancellation service, a bounded handoff transfers necessary facts without carrying discarded trial-and-error context; alongside this, an orchestrated workflow must not report overall success while a required step failed, satisfying the distinct requirements that a new agent gets only durable context from a completed job and that partial completion of a multi-step operation is handled safely.
Incorrect Answers
A:In an order-cancellation service, irrelevant turns can consume context and revive outdated assumptions and later actions can fail after an earlier side effect; the design would fail both conditions: a new agent gets only durable context from a completed job and partial completion of a multi-step operation is handled safely.
C:In an order-cancellation service, an orchestrated workflow must not report overall success while a required step failed, but irrelevant turns can consume context and revive outdated assumptions; therefore the separate requirement that a new agent gets only durable context from a completed job would remain unsatisfied.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 an order-cancellation service, a bounded handoff transfers necessary facts without carrying discarded trial-and-error context, but later actions can fail after an earlier side effect; therefore the separate requirement that partial completion of a multi-step operation is handled safely would remain unsatisfied.
Question 731
Before expanding a customer entitlement assistant, the owner must demonstrate that an out-of-policy request is escalated consistently, while also demonstrating that fast-changing information is checked against its publication date. Which design provides both controls?
- escalate only when the customer's sentiment is strongly negative; and record source timestamps and refresh time-sensitive evidence
- use an explicit exception trigger and designated human review queue; and record source timestamps and refresh time-sensitive evidence
- use an explicit exception trigger and designated human review queue; and reuse an old cached claim without its date
- escalate only when the customer's sentiment is strongly negative; and reuse an old cached claim without its date
- implement only use an explicit exception trigger and designated human review queue; make no corresponding change for the second requirement
Explanation
Correct Answer
B:In a customer entitlement assistant, policy-driven routing is predictable across changes in user tone; alongside this, freshness determines whether an otherwise credible fact applies now, satisfying the distinct requirements that an out-of-policy request is escalated consistently and that fast-changing information is checked against its publication date.
Incorrect Answers
A:In a customer entitlement assistant, freshness determines whether an otherwise credible fact applies now, but calmly phrased exceptions still exceed the agent's authority; therefore the separate requirement that an out-of-policy request is escalated consistently would remain unsatisfied.
C:In a customer entitlement assistant, policy-driven routing is predictable across changes in user tone, but stale evidence may no longer describe the current state; therefore the separate requirement that fast-changing information is checked against its publication date would remain unsatisfied.
D:In a customer entitlement assistant, calmly phrased exceptions still exceed the agent's authority and stale evidence may no longer describe the current state; the design would fail both conditions: an out-of-policy request is escalated consistently and fast-changing information is checked against its publication date.
E:In a customer entitlement assistant, policy-driven routing is predictable across changes in user tone, but the independent mechanism for the other criterion is absent: freshness determines whether an otherwise credible fact applies now; as written, the design does not ensure that fast-changing information is checked against its publication date.
Question 732
In a technical review of a cross-platform application migration, the architect must address the following constraints together: the assistant locates files containing a particular symbol or string; a destructive tool receives human authorization at execution time. Which change set is most defensible?
- use Grep to search file contents; and interrupt or deny the tool until an authorization callback approves it
- implement only use Grep to search file contents; make no corresponding change for the second requirement
- scan only filenames with Glob; and interrupt or deny the tool until an authorization callback approves it
- use Grep to search file contents; and collect a general approval at the start of the conversation
- scan only filenames with Glob; and collect a general approval at the start of the conversation
Explanation
Correct Answer
A:In a cross-platform application migration, text search locates occurrences efficiently before focused reading; alongside this, execution gates make the approval binding on the selected action and inputs, satisfying the distinct requirements that the assistant locates files containing a particular symbol or string and that a destructive tool receives human authorization at execution time.
Incorrect Answers
B:In a cross-platform application migration, text search locates occurrences efficiently before focused reading, but the independent mechanism for the other criterion is absent: execution gates make the approval binding on the selected action and inputs; as written, the design does not ensure that a destructive tool receives human authorization at execution time.
C:In a cross-platform application migration, execution gates make the approval binding on the selected action and inputs, but path matching does not inspect symbol occurrences inside files; therefore the separate requirement that the assistant locates files containing a particular symbol or string would remain unsatisfied.
D:In a cross-platform application migration, text search locates occurrences efficiently before focused reading, but blanket consent does not verify the later high-risk operation; therefore the separate requirement that a destructive tool receives human authorization at execution time would remain unsatisfied.
E:In a cross-platform application migration, path matching does not inspect symbol occurrences inside files and blanket consent does not verify the later high-risk operation; the design would fail both conditions: the assistant locates files containing a particular symbol or string and a destructive tool receives human authorization at execution time.
Question 733
Before expanding a source-grounded policy synthesis service, the owner must demonstrate that a failed MCP operation is visible as an error to the client, while also demonstrating that tool-call logging does not modify a successful tool result. Which design provides both controls?
- return a success-shaped value saying the operation probably worked; and rewrite every tool result as a free-form paraphrase
- mark the tool result as an error and supply structured failure details; and rewrite every tool result as a free-form paraphrase
- mark the tool result as an error and supply structured failure details; and use an observation-only hook for telemetry when rewriting is unnecessary
- return a success-shaped value saying the operation probably worked; and use an observation-only hook for telemetry when rewriting is unnecessary
Explanation
Correct Answer
C:In a source-grounded policy synthesis service, explicit error status separates failed execution from valid data content; alongside this, separating instrumentation from mutation reduces the risk of corrupting evidence, satisfying the distinct requirements that a failed MCP operation is visible as an error to the client and that tool-call logging does not modify a successful tool result.
Incorrect Answers
A:In a source-grounded policy synthesis service, downstream agents can mistake a failure for verified business data and summaries can discard fields that subsequent steps require; the design would fail both conditions: a failed MCP operation is visible as an error to the client and tool-call logging does not modify a successful tool result.
B:In a source-grounded policy synthesis service, explicit error status separates failed execution from valid data content, 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.
D:In a source-grounded policy synthesis service, separating instrumentation from mutation reduces the risk of corrupting evidence, but downstream agents can mistake a failure for verified business data; therefore the separate requirement that a failed MCP operation is visible as an error to the client would remain unsatisfied.
Question 734
An audit of a policy document normalization service identifies two unmet requirements: the model can use a tool result in its next decision and near-overlapping retrieval tools are selected by actual boundary. Select TWO actions that remedy those particular deficiencies.
- put the tool output only in application logs
- rely exclusively on short tool names for routing
- append the assistant tool-use response and matching user tool_result blocks to history
- document examples and conditions where each tool must not be used
Explanation
Correct Answers
C:For a policy document normalization service, Claude needs the prior request and the execution output in the conversation for the next inference, which directly establishes that the model can use a tool result in its next decision; the separate requirement that near-overlapping retrieval tools are selected by actual boundary is handled by a different action.
D:For a policy document normalization service, contrastive guidance helps tool choice when resource names look similar, which directly establishes that near-overlapping retrieval tools are selected by actual boundary; the separate requirement that the model can use a tool result in its next decision is handled by a different action.
Incorrect Answers
A:For a policy document normalization service, unseen logs cannot inform the model's next decision, so this option cannot establish that the model can use a tool result in its next decision; correctly addressing near-overlapping retrieval tools are selected by actual boundary elsewhere does not repair that deficiency.
B:For a policy document normalization service, names alone may not capture the distinction in scope or authorization, so this option cannot establish that near-overlapping retrieval tools are selected by actual boundary; correctly addressing the model can use a tool result in its next decision elsewhere does not repair that deficiency.
Question 735
An audit of an insurance document extraction agent asks how to make sure that conflicting model assessments of a disputed claim are reconciled against evidence. Which implementation should be selected?
- resolve the dispute against cited source evidence rather than majority vote
- combine independent review with schema and business-rule validation
- create a reconciliation stage that checks citations and resolves overlaps
- select whichever answer appears in the most generated responses
Explanation
Correct Answer
A:For an insurance document extraction agent, independent verification distinguishes a well-grounded minority view from shared guessing; this directly establishes that conflicting model assessments of a disputed claim are reconciled against evidence instead of relying on an unrelated workflow control.
Incorrect Answers
B:Although a critique may improve quality but cannot enforce machine-testable invariants by itself, that only supports the distinct goal that a high-risk result passes deterministic checks in addition to model critique, not the requirement that conflicting model assessments of a disputed claim are reconciled against evidence in an insurance document extraction agent.
C:Although synthesis must select mutually consistent claims with defensible evidence, that only supports the distinct goal that multiple draft branches are merged without contradictory statements, not the requirement that conflicting model assessments of a disputed claim are reconciled against evidence in an insurance document extraction agent.
D:In an insurance document extraction agent, agreement among models is not a substitute for factual grounding; as a result, the stated requirement that conflicting model assessments of a disputed claim are reconciled against evidence would still not be met.
Question 736
When designing a billing-dispute support agent, the requirement is not satisfied unless tool responses are associated with the exact invocations that produced them and important workflow state survives context compression or process restarts. Which implementation addresses the two independent conditions?
- return tool_result blocks using the original tool_use_id values; and depend on the model remembering a prior value forever
- return tool results with newly invented identifiers; and depend on the model remembering a prior value forever
- return tool results with newly invented identifiers; and persist authoritative task facts and artifact IDs outside ephemeral conversation text
- return tool_result blocks using the original tool_use_id values; and persist authoritative task facts and artifact IDs outside ephemeral conversation text
Explanation
Correct Answer
D:In a billing-dispute support agent, correlation identifiers bind asynchronous or parallel results to the right requests; alongside this, external durable state can be reloaded and checked when the session window changes, satisfying the distinct requirements that tool responses are associated with the exact invocations that produced them and that important workflow state survives context compression or process restarts.
Incorrect Answers
A:In a billing-dispute support agent, correlation identifiers bind asynchronous or parallel results to the right requests, but context limits and summarization can lose exact identifiers; therefore the separate requirement that important workflow state survives context compression or process restarts would remain unsatisfied.
B:In a billing-dispute support agent, the model cannot reliably associate a result with an outstanding call and context limits and summarization can lose exact identifiers; the design would fail both conditions: tool responses are associated with the exact invocations that produced them and important workflow state survives context compression or process restarts.
C:In a billing-dispute support agent, external durable state can be reloaded and checked when the session window changes, but the model cannot reliably associate a result with an outstanding call; therefore the separate requirement that tool responses are associated with the exact invocations that produced them would remain unsatisfied.
Question 737
The release gate for a purchase-order parsing service requires that reviewer changes remain auditable; it also requires that the coordinator can identify which subagent produced a broken artifact. Which TWO actions meet the specified checks? Choose TWO.
- overwrite an automated answer without preserving the reviewer decision
- log the reviewer identity, decision, justification, and artifact version
- propagate task IDs, error codes, and correlation references
- replace all nested errors with a generic coordinator failure
Explanation
Correct Answers
B:For a purchase-order parsing service, a versioned approval trail supports diagnosis and later governance reviews, which directly establishes that reviewer changes remain auditable; the separate requirement that the coordinator can identify which subagent produced a broken artifact is handled by a different action.
C:For a purchase-order parsing service, tracing execution lineage helps route the incident to the right failing component, which directly establishes that the coordinator can identify which subagent produced a broken artifact; the separate requirement that reviewer changes remain auditable is handled by a different action.
Incorrect Answers
A:For a purchase-order parsing service, the final state loses the reason it changed, so this option cannot establish that reviewer changes remain auditable; correctly addressing the coordinator can identify which subagent produced a broken artifact elsewhere does not repair that deficiency.
D:For a purchase-order parsing service, the original failing operation becomes opaque, so this option cannot establish that the coordinator can identify which subagent produced a broken artifact; correctly addressing reviewer changes remain auditable elsewhere does not repair that deficiency.
Question 738
For a technical due-diligence report generator, a reviewer identifies two independent obligations: file writes are blocked outside an approved workspace and the report is not drafted before evidence is checked. Which proposal should be approved?
- tell the model not to leave the workspace; and place evidence validation before the drafting stage
- implement only check resolved target paths in the pre-tool policy gate; make no corresponding change for the second requirement
- tell the model not to leave the workspace; and start drafting at the same time as the source search and assume success
- check resolved target paths in the pre-tool policy gate; and place evidence validation before the drafting stage
- check resolved target paths in the pre-tool policy gate; and start drafting at the same time as the source search and assume success
Explanation
Correct Answer
D:In a technical due-diligence report generator, a boundary check on the actual tool argument enforces the permitted filesystem scope; alongside this, a dependent stage must receive validated inputs to avoid propagating unsupported conclusions, satisfying the distinct requirements that file writes are blocked outside an approved workspace and that the report is not drafted before evidence is checked.
Incorrect Answers
A:In a technical due-diligence report generator, a dependent stage must receive validated inputs to avoid propagating unsupported conclusions, but language instructions do not intercept a disallowed filesystem operation; therefore the separate requirement that file writes are blocked outside an approved workspace would remain unsatisfied.
B:In a technical due-diligence report generator, a boundary check on the actual tool argument enforces the permitted filesystem scope, but the independent mechanism for the other criterion is absent: a dependent stage must receive validated inputs to avoid propagating unsupported conclusions; as written, the design does not ensure that the report is not drafted before evidence is checked.
C:In a technical due-diligence report generator, language instructions do not intercept a disallowed filesystem operation and the writer may treat unverified or absent evidence as real; the design would fail both conditions: file writes are blocked outside an approved workspace and the report is not drafted before evidence is checked.
E:In a technical due-diligence report generator, a boundary check on the actual tool argument enforces the permitted filesystem scope, 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.
Question 739
An incident postmortem for a static analysis automation requires controls so that few-shot material does not contain contradictory answer labels and that the reviewer does not simply echo the generator's confidence. Which paired remedy covers both failure modes?
- keep historical examples even after the labeling policy changes; and provide the generator's score as the sole review criterion
- keep historical examples even after the labeling policy changes; and give the reviewer the source and rubric, not only the draft's self-rating
- audit examples against the current decision rubric before use; and give the reviewer the source and rubric, not only the draft's self-rating
- audit examples against the current decision rubric before use; and provide the generator's score as the sole review criterion
Explanation
Correct Answer
C:In a static analysis automation, consistent labels reinforce the intended criteria rather than training ambiguity into the prompt; alongside this, independent evidence lets a critic challenge unsupported assertions, satisfying the distinct requirements that few-shot material does not contain contradictory answer labels and that the reviewer does not simply echo the generator's confidence.
Incorrect Answers
A:In a static analysis automation, stale demonstrations can contradict the updated standard and anchoring on a self-assigned grade hides actual defects; the design would fail both conditions: few-shot material does not contain contradictory answer labels and the reviewer does not simply echo the generator's confidence.
B:In a static analysis automation, independent evidence lets a critic challenge unsupported assertions, but stale demonstrations can contradict the updated standard; therefore the separate requirement that few-shot material does not contain contradictory answer labels would remain unsatisfied.
D:In a static analysis automation, consistent labels reinforce the intended criteria rather than training ambiguity into the prompt, but anchoring on a self-assigned grade hides actual defects; therefore the separate requirement that the reviewer does not simply echo the generator's confidence would remain unsatisfied.
Question 740
For a warranty claims assistant, a reviewer identifies two independent obligations: the system distinguishes acceptable and unacceptable edge cases and tool responses are associated with the exact invocations that produced them. Which proposal should be approved?
- specify inclusion criteria and explicit exclusion conditions; and return tool_result blocks using the original tool_use_id values
- provide only a broad topic label as the classification rule; and return tool_result blocks using the original tool_use_id values
- provide only a broad topic label as the classification rule; and return tool results with newly invented identifiers
- specify inclusion criteria and explicit exclusion conditions; and return tool results with newly invented identifiers
Explanation
Correct Answer
A:In a warranty claims assistant, negative boundaries teach the model when a near-match must not trigger; alongside this, correlation identifiers bind asynchronous or parallel results to the right requests, satisfying the distinct requirements that the system distinguishes acceptable and unacceptable edge cases and that tool responses are associated with the exact invocations that produced them.
Incorrect Answers
B:In a warranty claims assistant, correlation identifiers bind asynchronous or parallel results to the right requests, but generic labels leave ambiguous boundary cases undefined; therefore the separate requirement that the system distinguishes acceptable and unacceptable edge cases would remain unsatisfied.
C:In a warranty claims assistant, generic labels leave ambiguous boundary cases undefined and the model cannot reliably associate a result with an outstanding call; the design would fail both conditions: the system distinguishes acceptable and unacceptable edge cases and tool responses are associated with the exact invocations that produced them.
D:In a warranty claims assistant, negative boundaries teach the model when a near-match must not trigger, but the model cannot reliably associate a result with an outstanding call; therefore the separate requirement that tool responses are associated with the exact invocations that produced them would remain unsatisfied.