Question 350
A release must prove that users receive permission errors after successful package import will not recur. The application is used in this situation: An imported agent can authenticate in dev but not production. Which acceptance test most directly checks that failure mode?
- Test intended roles against each production-bound action
- Deploy a test agent and ensure discovery and ownership tracking
- Attempt a write from a read-only assistant
- Try a forbidden refund despite a permissive model instruction
Explanation
Correct Answer
A:The check is targeted rather than generic: production consent, identity and connection setup are distinct release checks; its pass/fail evidence applies to the workflow where an imported agent can authenticate in dev but not production and to the prior defect that users receive permission errors after successful package import.
Incorrect Answers
B:The suggested test serves business units create agents with no central visibility, in which security cannot identify which agents use protected connectors; the present scenario instead concerns the reported problem where users receive permission errors after successful package import.
C:Attempting a write from a read-only assistant verifies an adjacent objective where an agent should summarize account data but not alter it; it leaves untested the causal path in which production consent, identity and connection setup are distinct release checks for this case.
D:This check exercises a different system behavior: server-side access checks enforce real actor entitlements; a pass there cannot establish that users receive permission errors after successful package import no longer occurs in the required workflow.
Question 351
A solution architecture is being reviewed for this business situation: A support agent changes how it classifies priority intents. The observed defect is that high-severity tickets begin routing to general queues. Which architectural decision addresses the underlying requirement?
- Test Sales-to-Service customer context and case handoff
- Regression-test case routing after service agent changes
- Version agent instructions, tool definitions and model references together
- Use managed solutions for governed downstream environments
- Preserve customer communication approvals across solution releases
Explanation
Correct Answer
B:Without the intended behavior, high-severity tickets begin routing to general queues; the proposed architecture establishes the specific design principle that controlled service evaluations validate routing before rollout for the process where a support agent changes how it classifies priority intents.
Incorrect Answers
A:This design would make sense if a won opportunity should create a properly owned onboarding case, particularly where AI recommends an action but downstream case data is incomplete; the requested architecture instead needs the dependency that controlled service evaluations validate routing before rollout because high-severity tickets begin routing to general queues.
C:The technical behavior of this alternative is that a complete agent release manifest pins related runtime artifacts; the user case being assessed is a support agent changes how it classifies priority intents with a different failure: high-severity tickets begin routing to general queues.
D:The technical behavior of this alternative is that managed solution deployment supports controlled downstream customization boundaries; the user case being assessed is a support agent changes how it classifies priority intents with a different failure: high-severity tickets begin routing to general queues.
E:In the case where service agents draft customer messages but cannot send without review, the alternate design depends on business approval checks remain independent of agent version; it does not correct the target failure, high-severity tickets begin routing to general queues.
Question 352
A risk analysis highlights a threat in this workflow: Retraining uses revised labels and a different source snapshot. Exposure: A model release might combine training data with an unapproved candidate artifact. Which enforceable safeguard addresses it at the relevant boundary?
- Let authorized incident commanders override recommendations
- Block release when critical evidence is consistently omitted
- Prevent deployment of unapproved model/data combinations
- Retain document version and ACL information on every chunk
Explanation
Correct Answer
C:Because a model release might combine training data with an unapproved candidate artifact, a guardrail must constrain the actual system behavior: dataset lineage and model versioning support repeatable assessment; protecting a response alone cannot enforce that boundary.
Incorrect Answers
A:A specialized support team uses a nonstandard severity matrix; its guardrail addresses the possibility that automated severity suggestions might override an incident commander's decision; the present architecture must constrain the different threat that a model release might combine training data with an unapproved candidate artifact.
B:That protection is tied to the mechanism that labeled queries and expected evidence distinguish retrieval from synthesis; it does not prevent the distinct risk where a model release might combine training data with an unapproved candidate artifact in the system under review.
D:Another process could need this guardrail because chunked content might lose original permissions and version information; the design being assessed supports retraining uses revised labels and a different source snapshot and has the explicit exposure that a model release might combine training data with an unapproved candidate artifact.
Question 353
A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Retraining uses revised labels and a different source snapshot. Known defect: Model behavior changes without reproducible training provenance. Which KPI should be selected?
- Answer correctness on cross-section retrieval cases
- Grounding recall for critical enterprise questions
- Successful inferences within latency SLO
- Model versions with complete reproducibility evidence
Explanation
Correct Answer
D:Model versions with complete reproducibility evidence; this outcome measure belongs to the workflow where retraining uses revised labels and a different source snapshot; a change in it can help detect whether the defect that model behavior changes without reproducible training provenance is resolved.
Incorrect Answers
A:Answer correctness on cross-section retrieval cases; that indicator describes a distinct workflow where retrieved policy fragments lose context when documents are split; it does not quantify model versions with complete reproducibility evidence, which is needed for the stated case.
B:The alternative tracks grounding recall for critical enterprise questions, arising from the mechanism where labeled queries and expected evidence distinguish retrieval from synthesis; its movement alone says little about the failure that model behavior changes without reproducible training provenance.
C:That scorecard concerns the activity in which a specialized model must serve production apps reliably and its result of successful inferences within latency SLO; the reported shortcoming here is model behavior changes without reproducible training provenance.
Question 354
A product owner must measure whether an AI-enabled change actually improves this process: A stable specialized classification style persists despite prompt examples. The earlier problem was: Few-shot prompts require excessive tokens and yield inconsistent formats. Which outcome indicator is most relevant?
- Format compliance improvement on unseen tasks
- Cost per correct extraction under the response-time SLO
- Custom workflow completion against the required execution contract
- Schema-valid response rate
Explanation
Correct Answer
A:Because fine-tuning can adapt recurring behavior using vetted examples, the outcome worth following is format compliance improvement on unseen tasks; the metric is tied to the reported issue—few-shot prompts require excessive tokens and yield inconsistent formats—for the scenario where a stable specialized classification style persists despite prompt examples.
Incorrect Answers
B:High-volume narrow extraction needs predictable latency and cost; its intended measurement is cost per correct extraction under the response-time SLO; substituting that result here would not capture whether few-shot prompts require excessive tokens and yield inconsistent formats has been corrected.
C:A change in custom workflow completion against the required execution contract is informative for the solution demands its own model selection and hosting policies; it does not establish whether the current process achieves format compliance improvement on unseen tasks.
D:The premise of this indicator is clear instructions and structured outputs reduce avoidable ambiguity; the architecture being evaluated instead serves a stable specialized classification style persists despite prompt examples, and its success is measured by format compliance improvement on unseen tasks.
Question 355
A release must prove that a prompt change during close generates unexplained exceptions will not recur. The application is used in this situation: Invoice posting agents affect critical close processes. Which acceptance test most directly checks that failure mode?
- Try accessing a restricted training corpus from an unrelated project
- Simulate rollback during a nonproduction close cycle
- Upgrade one app while retaining the prior dependent version
- Execute both approved and rejected concession scenarios
Explanation
Correct Answer
B:Simulating rollback during a nonproduction close cycle is aligned with the failure that a prompt change during close generates unexplained exceptions; its causal relevance is that coordinated release windows account for transaction criticality for the situation where invoice posting agents affect critical close processes.
Incorrect Answers
A:A pass on that verification would speak to proprietary records are copied to an open test workspace; it would not show that coordinated release windows account for transaction criticality, which is needed to avoid the observed defect a prompt change during close generates unexplained exceptions.
C:This check exercises a different system behavior: cross-app release plans validate shared contracts and timing; a pass there cannot establish that a prompt change during close generates unexplained exceptions no longer occurs in the required workflow.
D:This check exercises a different system behavior: transaction-path tests confirm recommendation, review and posting records; a pass there cannot establish that a prompt change during close generates unexplained exceptions no longer occurs in the required workflow.
Question 356
A business stakeholder describes this operating situation: A helpful assistant can still produce unsafe financial or legal guidance. Testing instead shows that pilot satisfaction hides severe low-frequency failures. What design approach is most appropriate?
- Implement harm-prevention testing before broader adoption
- Prioritize the backlog by impact, recurrence and affected audience
- Cluster similar agent failures using embeddings or AI-assisted analysis
- Design meaningful human oversight for consequential recommendations
Explanation
Correct Answer
A:The architecture must reconcile two facts: a helpful assistant can still produce unsafe financial or legal guidance, yet pilot satisfaction hides severe low-frequency failures; the selected pattern does so through the mechanism that risk-based red-team evaluations identify foreseeable harm.
Incorrect Answers
B:Risk-weighted backlog scoring includes severity and exposure; its main purpose is product owners favor easy fixes over high-consequence failures, not correction of the reported defect where pilot satisfaction hides severe low-frequency failures in the target process.
C:In the case where thousands of transcripts contain recurring misunderstandings, the alternate design depends on semantic clustering groups related failure patterns for review; it does not correct the target failure, pilot satisfaction hides severe low-frequency failures.
D:There is a valid place for the mechanism that explainable review interfaces support accountable judgment, namely AI proposes eligibility decisions affecting customers; it does not remove the specific defect that pilot satisfaction hides severe low-frequency failures from this case.
Question 357
Security and business owners want to keep this capability: A user asks the agent to read an unrelated executive record. They must also prevent the following: A persuasive jailbreak might cause the backend to reveal privileged records. Which guardrail should the design incorporate?
- Provide accessible human escalation paths
- Require approval and a separate action identity for mutations
- Deny external sends unless independently approved
- Reject requests independent of instruction wording
Explanation
Correct Answer
D:Avoiding the consequence where a persuasive jailbreak might cause the backend to reveal privileged records requires a control for a user asks the agent to read an unrelated executive record; in this architecture, authorization evaluations verify server-side deny behavior, which is the point at which the boundary can be applied.
Incorrect Answers
A:This restriction is associated with chat and voice agents have different interaction patterns and addresses the exposure that voice users might be abandoned without an accessible path to a human; it does not specifically protect the target workflow against a persuasive jailbreak might cause the backend to reveal privileged records.
B:The alternate guard addresses a read-only assistant might acquire write operations by sharing a tool identity for an agent should summarize account data but not alter it; the present threat, a persuasive jailbreak might cause the backend to reveal privileged records, needs a different enforcement boundary.
C:Another process could need this guardrail because an injected tool instruction might export customer records externally; the design being assessed supports a user asks the agent to read an unrelated executive record and has the explicit exposure that a persuasive jailbreak might cause the backend to reveal privileged records.
Question 358
Security and business owners want to keep this capability: AI assistance reads operational records through Dataverse. They must also prevent the following: AI retrieval might expose virtual finance entities outside the reader's role. Which guardrail should the design incorporate?
- Rotate compromised keys and enforce least-privilege callers
- Limit entity visibility to authorized finance roles
- Do not approve procurement while financial hold status is unknown
- Restrict production to trusted vetted model versions
Explanation
Correct Answer
B:In the scenario where AI assistance reads operational records through Dataverse, the explicit risk is that AI retrieval might expose virtual finance entities outside the reader's role; enforcement must follow the mechanism by which integration tests protect virtual-entity schema and source semantics.
Incorrect Answers
A:A sensitive model endpoint is reachable using a broadly shared key; its guardrail addresses the possibility that a leaked endpoint credential might permit inference from an untrusted client; the present architecture must constrain the different threat that AI retrieval might expose virtual finance entities outside the reader's role.
C:Another process could need this guardrail because a procurement approval might proceed before a finance hold is resolved; the design being assessed supports AI assistance reads operational records through Dataverse and has the explicit exposure that AI retrieval might expose virtual finance entities outside the reader's role.
D:Deployment automation pulls model artifacts from unverified origins; its guardrail addresses the possibility that an attacker might swap a model artifact in the deployment pipeline; the present architecture must constrain the different threat that AI retrieval might expose virtual finance entities outside the reader's role.
Question 359
A staging test has to detect the specific defect that teams learn about problems only when someone asks the agent. In the workflow where shipment exceptions should initiate investigation without a chat message, which exercise verifies the corrective behavior?
- Emit a sample exception event and inspect one workflow start
- Ask for a variance explanation with incomplete data
- Test edge cases with conflicting documents and incomplete facts
- Open the page for different roles and verify context limits
Explanation
Correct Answer
A:Emiting a sample exception event and inspecting one workflow start challenges the path where an event trigger starts the bounded agent workflow on state change for the business setting shipment exceptions should initiate investigation without a chat message; the specific defect to exclude is that teams learn about problems only when someone asks the agent.
Incorrect Answers
B:The suggested test serves finance asks why a variance was flagged, in which answers assert causes without referencing the ledger; the present scenario instead concerns the reported problem where teams learn about problems only when someone asks the agent.
C:A pass on that verification would speak to procurement exceptions require comparing several policies and records; it would not show that an event trigger starts the bounded agent workflow on state change, which is needed to avoid the observed defect teams learn about problems only when someone asks the agent.
D:In the alternate setting where staff need a contextual workspace assembled around one customer case, the test exercises generated interface components can be grounded in governed app data and defined actions; even a pass would not show that teams learn about problems only when someone asks the agent has been prevented.
Question 360
A product owner must measure whether an AI-enabled change actually improves this process: Staff need a contextual workspace assembled around one customer case. The earlier problem was: Generic chat loses visibility into related records and actions. Which outcome indicator is most relevant?
- Unauthorized data fields sent to model actions
- Dependent applications using the approved AI capability version
- Ticket updates completed with attributable credentials
- Case-workspace completion without navigation detours
- User understanding of verified versus suggested content
Explanation
Correct Answer
D:This scorecard needs an outcome related to staff need a contextual workspace assembled around one customer case; case-workspace completion without navigation detours. That indicator can be checked against the earlier symptom that generic chat loses visibility into related records and actions.
Incorrect Answers
A:A canvas app passes customer records to a model action; its intended measurement is unauthorized data fields sent to model actions; substituting that result here would not capture whether generic chat loses visibility into related records and actions has been corrected.
B:For the workflow where multiple low-code apps need a common extraction prompt, dependent applications using the approved AI capability version is an actionable measure; it does not establish whether generic chat loses visibility into related records and actions in the situation under review.
C:The proposed metric concerns an agent needs to update a governed ticket status, whereas the failure in this case is that generic chat loses visibility into related records and actions; the indicator needed here is case-workspace completion without navigation detours.
E:The alternative tracks user understanding of verified versus suggested content, arising from the mechanism where visual provenance separates authoritative records from generated advice; its movement alone says little about the failure that generic chat loses visibility into related records and actions.