Question 330
The operating team must monitor whether an architectural correction works. Business context: A backend agent must call an approved Azure service without embedded secrets. Historical symptom: Long-lived API keys leak from configuration and diagnostics. Which outcome measure is the best match?
- Production agent components using approved workload identity
- Privileged mutations without verified principal authorization
- Unauthorized inference requests successfully blocked
- Authorized tool calls succeeding after promotion
Explanation
Correct Answer
A:For the business use where a backend agent must call an approved Azure service without embedded secrets, the result to quantify is production agent components using approved workload identity; it connects measurement to the mechanism where managed identity provides credential-less workload authentication where supported, not merely to agent traffic.
Incorrect Answers
B:Privileged mutations without verified principal authorization; that indicator describes a distinct workflow where prompt wording persuades the agent to submit a privileged operation; it does not quantify production agent components using approved workload identity, which is needed for the stated case.
C:A change in unauthorized inference requests successfully blocked is informative for a sensitive model endpoint is reachable using a broadly shared key; it does not establish whether the current process achieves production agent components using approved workload identity.
D:The premise of this indicator is production consent, identity and connection setup are distinct release checks; the architecture being evaluated instead serves a backend agent must call an approved Azure service without embedded secrets, and its success is measured by production agent components using approved workload identity.
Question 331
The architecture board has evidence that candidate improves one metric while violating another. The affected situation is: Teams promote a custom model because it feels better. What change most directly establishes the missing behavior?
- Validate custom extraction models by field and document class
- Set acceptance thresholds and reproducible baselines
- Build risk-based test coverage for critical user journeys
- Promote grounding datasets through quality gates
Explanation
Correct Answer
B:The architecture must reconcile two facts: teams promote a custom model because it feels better, yet candidate improves one metric while violating another; the selected pattern does so through the mechanism that predefined gates weigh accuracy, latency and business risk.
Incorrect Answers
A:For a document model performs well overall but fails on tax identifiers, this other approach relies on the fact that stratified field metrics protect high-impact values; the issue here is candidate improves one metric while violating another; that mechanism does not establish the required predefined gates weigh accuracy, latency and business risk.
C:In the case where agent supports refunds, order status and generic FAQs, the alternate design depends on risk-weighting prioritizes high-impact actions and failure paths; it does not correct the target failure, candidate improves one metric while violating another.
D:In the case where new documents enter a production source automatically, the alternate design depends on ingestion gates check classification, freshness and ownership; it does not correct the target failure, candidate improves one metric while violating another.
Question 332
A policy review covers an agent for which a stable specialized classification style persists despite prompt examples. Its most relevant exposure is that sensitive or unlicensed content could be incorporated into tuning examples. What safeguard should be required before release?
- Operate independent identity, safety and observability controls
- Require human review for high-impact low-confidence classifications
- Exclude secrets and ensure training-data rights
- Reject malformed output before writing business records
Explanation
Correct Answer
C:Avoiding the consequence where sensitive or unlicensed content could be incorporated into tuning examples requires a control for a stable specialized classification style persists despite prompt examples; in this architecture, fine-tuning can adapt recurring behavior using vetted examples, which is the point at which the boundary can be applied.
Incorrect Answers
A:The solution demands its own model selection and hosting policies; its guardrail addresses the possibility that an owned custom runtime might lack the host's identity and safety governance; the present architecture must constrain the different threat that sensitive or unlicensed content could be incorporated into tuning examples.
B:The safeguard for the classifier confuses complaints with refund requests deals with a different threat: a low-confidence label might be used for an irreversible eligibility decision; even after it is enforced, the exposure that sensitive or unlicensed content could be incorporated into tuning examples remains possible in the target workflow where fine-tuning can adapt recurring behavior using vetted examples.
D:For the use case where agents return inconsistent fields for downstream integrations, the control addresses the threat that invalid generated JSON might be written directly into a business record; the specified agent must be protected against sensitive or unlicensed content could be incorporated into tuning examples.
Question 333
An agent-assisted service does not behave as specified: The investment case overstates value by ignoring review and hosting. It is used in this situation: A pilot reports gross labor savings but omits operating expenses. Which observable diagnostic procedure is best?
- Trace approval delays to unowned architectural decisions
- Score use cases against data, identity, governance and integration readiness
- Reconcile model assumptions against invoices and actual time studies
- Benchmark accuracy, throughput and resource demands together
- Audit rollout settings and incident timelines for uncontrolled changes
Explanation
Correct Answer
C:Where the investment case overstates value by ignoring review and hosting, the useful evidence comes from reconciling model assumptions against invoices and actual time studies; its focus matches the architectural mechanism that net benefit compares realized outcomes with implementation and recurring costs in the process where a pilot reports gross labor savings but omits operating expenses.
Incorrect Answers
A:Tracing approval delays to unowned architectural decisions explores a separate connection, where a named strategy team assigns decision rights and resolves cross-function tradeoffs; the diagnosed requirement is a pilot reports gross labor savings but omits operating expenses, and the actual defect is the investment case overstates value by ignoring review and hosting.
B:The first-wave pilot requests broad access to unreliable records; examining that other path does not determine whether net benefit compares realized outcomes with implementation and recurring costs, which is what distinguishes the issue in the stated use case.
D:A general large model exceeds per-document unit economics; examining that other path does not determine whether net benefit compares realized outcomes with implementation and recurring costs, which is what distinguishes the issue in the stated use case.
E:The proposed inspection is associated with new behavior must be evaluated without exposing the full workforce, where controlled release gates restrict blast radius and support rollback; it would not expose the failure mode that the investment case overstates value by ignoring review and hosting with the required specificity.
Question 334
A design adds autonomous behavior to a process where sales copilots should not expose sensitive margin assumptions. This creates a documented risk: An ERP connector might expose protected margin columns in its response payload. Which enforced control is the best fit?
- Block unclassified high-risk fields at the connector boundary
- Fail closed on unresolved caller identity
- Assign glossary approval to service process owners
- Require least-privilege service permissions and action confirmation
Explanation
Correct Answer
A:Because an ERP connector might expose protected margin columns in its response payload, a guardrail must constrain the actual system behavior: field projection and endpoint authorization minimize exposed data; protecting a response alone cannot enforce that boundary.
Incorrect Answers
B:The alternate boundary mitigates an anonymous caller might execute an account mutation after providing only a guessed ID, which is relevant to anonymous callers can ask generic questions but not change orders; it leaves the current workflow vulnerable to an ERP connector might expose protected margin columns in its response payload.
C:The alternate guard addresses unreviewed synonyms could redefine legally significant customer-service classifications for agents must distinguish an incident, complaint, inquiry and entitlement; the present threat, an ERP connector might expose protected margin columns in its response payload, needs a different enforcement boundary.
D:The alternate guard addresses a tool might update tickets using a shared identity with too many permissions for an agent needs to update a governed ticket status; the present threat, an ERP connector might expose protected margin columns in its response payload, needs a different enforcement boundary.
Question 335
Risk reviewers need a control, not merely a monitoring dashboard. The use case is: A new model reduces price but also degrades exact extractions. Potential harm: Lower-cost model routing might undermine critical extraction accuracy. Which protection fits?
- Limit retroactive alteration of approved QA evidence
- Refuse cost optimization that breaches business-quality thresholds
- Roll back if privileged actions appear in unexpected paths
- Enforce owner sign-off for high-risk grounding collections
Explanation
Correct Answer
B:Avoiding the consequence where lower-cost model routing might undermine critical extraction accuracy requires a control for a new model reduces price but also degrades exact extractions; in this architecture, joint quality and consumption measures capture real efficiency, which is the point at which the boundary can be applied.
Incorrect Answers
A:The alternate boundary mitigates old evaluation evidence might be overwritten to conceal a failed release gate, which is relevant to an updated agent performs differently months after release; it leaves the current workflow vulnerable to lower-cost model routing might undermine critical extraction accuracy.
C:This restriction is associated with a new topic description changes which agent is invoked and addresses the exposure that a new orchestration release might suddenly include unexpected privileged tool paths; it does not specifically protect the target workflow against lower-cost model routing might undermine critical extraction accuracy.
D:The alternate boundary mitigates sensitive ownerless documents might enter production grounding, which is relevant to new documents enter a production source automatically; it leaves the current workflow vulnerable to lower-cost model routing might undermine critical extraction accuracy.
Question 336
After deployment, stakeholders need an interpretable measure for this use case: A document model performs well overall but fails on tax identifiers. The pilot showed that a high aggregate score masks critical field errors. Which metric captures the needed result?
- Worst-performing segment error rate
- Critical-field extraction accuracy across document classes
- Critical releases with reproducible evaluation and sign-off records
- Unsupported claims emitted in low-evidence tests
Explanation
Correct Answer
B:Because stratified field metrics protect high-impact values, the outcome worth following is critical-field extraction accuracy across document classes; the metric is tied to the reported issue—a high aggregate score masks critical field errors—for the scenario where a document model performs well overall but fails on tax identifiers.
Incorrect Answers
A:For the workflow where a classifier underperforms on one language group, worst-performing segment error rate is an actionable measure; it does not establish whether a high aggregate score masks critical field errors in the situation under review.
C:The proposed metric concerns an updated agent performs differently months after release, whereas the failure in this case is that a high aggregate score masks critical field errors; the indicator needed here is critical-field extraction accuracy across document classes.
D:The alternative tracks unsupported claims emitted in low-evidence tests, arising from the mechanism where evidence and refusal evaluations enforce source-supported claims; its movement alone says little about the failure that a high aggregate score masks critical field errors.
Question 337
An enterprise team is making a platform design decision. Business context: Support tickets reveal an unusual multistep conversation failure. Present failure: The fixed scenario is not added to future test suites. Which approach resolves the mismatch?
- Automate repeatable evaluation gates before publishing
- Use production feedback to propose targeted regression tests
- Generate negative cases from tool and identity specifications
- Link negative feedback to grounded traces and source versions
Explanation
Correct Answer
B:The defect is that the fixed scenario is not added to future test suites; the expected business case is that support tickets reveal an unusual multistep conversation failure; the design fits because feedback-derived examples convert incidents into lasting safeguards.
Incorrect Answers
A:In the case where prompt revisions cause intermittent regressions, the alternate design depends on stable automated suites compare response quality across versions; it does not correct the target failure, the fixed scenario is not added to future test suites.
C:Happy-path fixtures omit unauthorized and invalid inputs; the alternate design can address that situation because specification-guided generation covers missing fields and denied roles, whereas this question needs a solution to the fixed scenario is not added to future test suites.
D:The technical behavior of this alternative is that trace and content-version linkage makes failures testable; the user case being assessed is support tickets reveal an unusual multistep conversation failure with a different failure: the fixed scenario is not added to future test suites.
Question 338
An enterprise team is making a platform design decision. Business context: Independent teams choose inconsistent evaluation and security methods. Present failure: One division releases agents without risk review. Which approach resolves the mismatch?
- Maintain a reusable approved-agent and prompt catalogue
- Publish discoverable data contracts with schema versions
- Use A2A for agent-to-agent collaboration across runtimes
- Establish a cross-functional AI Center of Excellence playbook
Explanation
Correct Answer
D:The architecture must reconcile two facts: independent teams choose inconsistent evaluation and security methods, yet one division releases agents without risk review; the selected pattern does so through the mechanism that a CoE publishes patterns, reusable controls, standards and escalation paths.
Incorrect Answers
A:For another type of architecture, discovery and ownership metadata support reuse and controlled deprecation is useful; this design must instead support independent teams choose inconsistent evaluation and security methods despite the problem where one division releases agents without risk review.
B:For multiple AI teams consume a changing source API, this other approach relies on the fact that versioned schemas and compatibility policy make interfaces predictable; the issue here is one division releases agents without risk review; that mechanism does not establish the required a CoE publishes patterns, reusable controls, standards and escalation paths.
C:When independently hosted agents must exchange tasks and results, an agent interoperability protocol supports explicit task delegation between agents; in the stated workflow, however, one division releases agents without risk review; the relevant architectural behavior is that a CoE publishes patterns, reusable controls, standards and escalation paths.
Question 339
A quality engineer is writing a regression case after the incident where no log identifies which source version was indexed. The underlying business scenario is: Agents begin citing an obsolete article after an ingestion job. What should the test exercise?
- Submit a malicious mislabeled training record and inspect review
- Reindex a modified document and verify traceability
- Roll back a staged faulty transformation and compare output
- Query identical terms as HR and non-HR users
Explanation
Correct Answer
B:Reindexing a modified document and verifying traceability is aligned with the failure that no log identifies which source version was indexed; its causal relevance is that content lineage records update, transformation and index revision for the situation where agents begin citing an obsolete article after an ingestion job.
Incorrect Answers
A:In a solution where business analysts can submit examples but not change production behavior, submitting a malicious mislabeled training record and inspecting review could catch the defect that unvetted labeled data poisons a retrained classifier; the current release gate must challenge the failure where no log identifies which source version was indexed.
C:In the alternate setting where a preprocessing release changes category labels incorrectly, the test exercises versioned data transformations and model artifacts enable consistent rollback; even a pass would not show that no log identifies which source version was indexed has been prevented.
D:The verification concerning users have different rights to the same knowledge corpus checks retrieval applies effective source entitlements to candidate documents against agent cites a private HR document to unauthorized staff; its result says little about the observed issue where no log identifies which source version was indexed.
Question 340
Troubleshooting starts from a business complaint: Happy-path fixtures omit unauthorized and invalid inputs. The affected process is: Agents can call APIs with customer IDs and amount values. Which investigation is most directly tied to the reported defect?
- Compare generated tests to authentication and schema contracts
- Inspect conversation state and parameter mapping across turns
- Reconcile audit IDs across each application
- Inspect test distribution against business consequence tiers
Explanation
Correct Answer
A:For the case where agents can call APIs with customer IDs and amount values, the diagnostic path must explain the failure that happy-path fixtures omit unauthorized and invalid inputs; comparing generated tests to authentication and schema contracts targets the mechanism by which specification-guided generation covers missing fields and denied roles.
Incorrect Answers
B:The proposed inspection is associated with agent must remember order context while asking for missing details, where multi-turn tests assess state tracking, entity separation and follow-up questions; it would not expose the failure mode that happy-path fixtures omit unauthorized and invalid inputs with the required specificity.
C:In the separate scenario where service recommends a price adjustment requiring finance approval, the same investigation could explain the final order lacks traceability to the AI suggestion and approver; it does not isolate the reported cause in a system where agents can call APIs with customer IDs and amount values.
D:The test suite spends most cases on low-risk FAQs; examining that other path does not determine whether specification-guided generation covers missing fields and denied roles, which is what distinguishes the issue in the stated use case.