Question 310
Release reviewers will not accept a generic success message. The business scenario is that data scientists cannot explain why a newer classifier behaves differently, and the historical failure is: Models move to production without an audit of training inputs. Which test is appropriately targeted?
- Reconstruct one released model's source and threshold decisions
- Evaluate candidate and baseline on identical frozen examples
- Execute an authorized and a denied deletion in staging
- Degrade sample images and test extraction stability
Explanation
Correct Answer
A:Reconstructing one released model's source and threshold decisions is aligned with the failure that models move to production without an audit of training inputs; its causal relevance is that a model registry ties versions to lineage, evaluations and approval for the situation where data scientists cannot explain why a newer classifier behaves differently.
Incorrect Answers
B:The tested behavior is that predefined gates weigh accuracy, latency and business risk; the necessary evidence here concerns a model registry ties versions to lineage, evaluations and approval and the previously observed models move to production without an audit of training inputs.
C:A pass on that verification would speak to a tool deletes a record following an ambiguous conversation; it would not show that a model registry ties versions to lineage, evaluations and approval, which is needed to avoid the observed defect models move to production without an audit of training inputs.
D:For production documents differ from clean training samples, the alternate test is about whether stress tests measure sensitivity to expected input variation; it cannot prove the outcome needed after the reported regression where models move to production without an audit of training inputs.
Question 311
Security and business owners want to keep this capability: Independently hosted agents must exchange tasks and results. They must also prevent the following: An external agent might accept delegated tasks from a spoofed sender. Which guardrail should the design incorporate?
- Restrict access through source permissions, not prompt wording
- Honor cross-app role permissions and data minimization
- Validate sending-agent identity and delegated authorization
- Confirm tenant policy and data access before enabling adoption
Explanation
Correct Answer
C:The business setting is independently hosted agents must exchange tasks and results, with the exposure that an external agent might accept delegated tasks from a spoofed sender; the enforceable protection must correspond to the path where an agent interoperability protocol supports explicit task delegation between agents.
Incorrect Answers
A:The safeguard for employee policies reside in governed Microsoft 365 documents deals with a different threat: an agent prompt could claim permission to read a private SharePoint document; even after it is enforced, the exposure that an external agent might accept delegated tasks from a spoofed sender remains possible in the target workflow where an agent interoperability protocol supports explicit task delegation between agents.
B:The safeguard for sales and service agents need the same customer and case context deals with a different threat: cross-application aggregation might expose records outside the user's Dynamics permissions; even after it is enforced, the exposure that an external agent might accept delegated tasks from a spoofed sender remains possible in the target workflow where an agent interoperability protocol supports explicit task delegation between agents.
D:The alternate boundary mitigates a standard agent might violate tenant access policy when activated for many users, which is relevant to users need ordinary meeting summaries and follow-up drafts; it leaves the current workflow vulnerable to an external agent might accept delegated tasks from a spoofed sender.
Question 312
A policy review covers an agent for which finance asks why a variance was flagged. Its most relevant exposure is that an analytical response might present an unsupported hypothesis as a ledger fact. What safeguard should be required before release?
- Require least-privilege service permissions and action confirmation
- Block release when critical evidence is consistently omitted
- Reject generated fields outside the approved schema
- Label unsupported explanations as hypotheses or abstain
Explanation
Correct Answer
D:In the scenario where finance asks why a variance was flagged, the explicit risk is that an analytical response might present an unsupported hypothesis as a ledger fact; enforcement must follow the mechanism by which grounded explanatory responses distinguish evidence from hypothesis.
Incorrect Answers
A:The proposed control targets a tool might update tickets using a shared identity with too many permissions, while the given risk scenario is an analytical response might present an unsupported hypothesis as a ledger fact; these exposures arise in different operational contexts.
B:The safeguard for stakeholders report hallucinations but cannot isolate search errors deals with a different threat: a release might proceed despite consistently missing the required evidence passage; even after it is enforced, the exposure that an analytical response might present an unsupported hypothesis as a ledger fact remains possible in the target workflow where grounded explanatory responses distinguish evidence from hypothesis.
C:Where an agent must turn service notes into a fixed JSON summary, the guard can address generated response keys might be trusted even when they violate the schema; the exposure that an analytical response might present an unsupported hypothesis as a ledger fact remains relevant to the present assessment.
Question 313
A quality engineer is writing a regression case after the incident where a document index cannot answer current case ownership reliably. The underlying business scenario is: Users ask about governed CRM records rather than prose manuals. What should the test exercise?
- Ask for a case field and verify the exact authorized record
- Replace a source document without training and verify updated answers
- Seed lookalike records and check correct entity separation
- Add a confidential draft and verify it is excluded for unauthorized users
Explanation
Correct Answer
A:For the required workflow where users ask about governed CRM records rather than prose manuals, the test needs to reproduce or prevent the failure where a document index cannot answer current case ownership reliably; asking for a case field and verifying the exact authorized record exercises the path where Dataverse grounding exposes appropriate structured records to the agent.
Incorrect Answers
B:This check exercises a different system behavior: retrieval separates volatile knowledge from model parameters; a pass there cannot establish that a document index cannot answer current case ownership reliably no longer occurs in the required workflow.
C:In a solution where multiple systems use inconsistent customer identifiers and duplicate cases, seeding lookalike records and checking correct entity separation could catch the defect that responses merge two unrelated customers with similar names; the current release gate must challenge the failure where a document index cannot answer current case ownership reliably.
D:The verification concerning search spans internal manuals with different owners and permissions checks metadata enables filtering, review, retention and traceable answers against a model retrieves a draft manual without any accountable owner; its result says little about the observed issue where a document index cannot answer current case ownership reliably.
Question 314
A product owner must measure whether an AI-enabled change actually improves this process: A sensitive model endpoint is reachable using a broadly shared key. The earlier problem was: Untrusted clients submit requests and exfiltrate outputs. Which outcome indicator is most relevant?
- Unauthorized inference requests successfully blocked
- Model releases containing unapproved training records
- Cross-role cache disclosure incidents
- Unauthorized accesses to model-training datasets
Explanation
Correct Answer
A:For the business use where a sensitive model endpoint is reachable using a broadly shared key, the result to quantify is unauthorized inference requests successfully blocked; it connects measurement to the mechanism where identity-based access and supported network isolation reduce exposure, not merely to agent traffic.
Incorrect Answers
B:Although model releases containing unapproved training records is observable, the affected process involves a sensitive model endpoint is reachable using a broadly shared key; its acceptance outcome is unauthorized inference requests successfully blocked, not the alternative measurement.
C:For the situation where shared result caching makes answers fast across roles, cross-role cache disclosure incidents is a sensible indicator; the problem that untrusted clients submit requests and exfiltrate outputs requires measurement of unauthorized inference requests successfully blocked.
D:For the situation where proprietary records are copied to an open test workspace, unauthorized accesses to model-training datasets is a sensible indicator; the problem that untrusted clients submit requests and exfiltrate outputs requires measurement of unauthorized inference requests successfully blocked.
Question 315
During integration testing, a separate application breaks established productivity workflows. The associated business context is: Staff need a role-specific conversation within Copilot and Teams. Which architectural remedy is the strongest match?
- Choose a declarative Microsoft 365 Copilot agent for in-host work
- Use explicit handoff contracts for asynchronous specialist work
- Buy a supported managed capability for standard requirements
- Use a supervisor agent that routes to specialized child agents
- Embed a Copilot-powered interaction inside an existing line-of-business app
Explanation
Correct Answer
A:Staff need a role-specific conversation within Copilot and Teams; its failing behavior is that a separate application breaks established productivity workflows; applying the underlying design principle—declarative agents use host orchestration with tailored instructions and actions—connects the requirement to an implementable control.
Incorrect Answers
B:When downstream research agents run longer than the interactive request, task IDs, status transitions, deadlines and durable handoff state support coordination; in the stated workflow, however, a separate application breaks established productivity workflows; the relevant architectural behavior is that declarative agents use host orchestration with tailored instructions and actions.
C:This design would make sense if the requirement is common and product coverage is adequate, particularly where a bespoke solution creates needless infrastructure overhead; the requested architecture instead needs the dependency that declarative agents use host orchestration with tailored instructions and actions because a separate application breaks established productivity workflows.
D:One general agent conflates expertise and may call the wrong tools; the alternate design can address that situation because a coordinating agent selects bounded specialists and merges their outputs, whereas this question needs a solution to a separate application breaks established productivity workflows.
E:This design would make sense if users must remain in a proprietary work-order application, particularly where requiring a separate chat portal increases context switching; the requested architecture instead needs the dependency that declarative agents use host orchestration with tailored instructions and actions because a separate application breaks established productivity workflows.
Question 316
Release reviewers will not accept a generic success message. The business scenario is that sales users say deal stage differently across regions, and the historical failure is: An agent updates the wrong pipeline field from an informal phrase. Which test is appropriately targeted?
- Test duplicate legal names with different external IDs
- Change a permitted credit field and verify updated authorized retrieval
- Request stage updates using local terminology and verify the target field
- Query as multiple roles and confirm only authorized fields appear
Explanation
Correct Answer
C:The scenario is that sales users say deal stage differently across regions; the defect to rule out is that an agent updates the wrong pipeline field from an informal phrase; testing via requesting stage updates using local terminology and verifying the target field challenges the design dependency where canonical business terminology maps language to underlying Dynamics records.
Incorrect Answers
A:For two systems share company names but not primary keys, the alternate test is about whether a mapping service resolves stable identities before enrichment; it cannot prove the outcome needed after the reported regression where an agent updates the wrong pipeline field from an informal phrase.
B:This check exercises a different system behavior: a supported connector can retrieve external information with explicit identity mapping; a pass there cannot establish that an agent updates the wrong pipeline field from an informal phrase no longer occurs in the required workflow.
D:A pass on that verification would speak to sales copilots should not expose sensitive margin assumptions; it would not show that canonical business terminology maps language to underlying Dynamics records, which is needed to avoid the observed defect an agent updates the wrong pipeline field from an informal phrase.
Question 317
The fault affects an AI-powered workflow in which a Teams user should query and update a supported external service. Specifically, read-only knowledge alone cannot create a service request. Which diagnostic investigation best distinguishes the relevant failure path?
- Review confusion pairs against the customer-service glossary
- Review action inputs, examples and schema-parse logs
- Inspect plugin capabilities and authentication configuration
- Compare source timestamps and principal mapping on connector calls
Explanation
Correct Answer
C:Read-only knowledge alone cannot create a service request; to attribute the defect rather than guess at a fix, inspecting plugin capabilities and authentication configuration examines the dependency where a plugin provides authorized callable tools to an agent.
Incorrect Answers
A:In the separate scenario where agents must distinguish an incident, complaint, inquiry and entitlement, the same investigation could explain Copilot misclassifies a complaint as an ordinary information request; it does not isolate the reported cause in a system where a Teams user should query and update a supported external service.
B:An agent must turn service notes into a fixed JSON summary; its matching diagnostic check is reasonable for a general conversational turn produces inconsistent output fields; the present investigation needs evidence for the different failure where read-only knowledge alone cannot create a service request.
D:For the failure that Copilot relies on last quarter's manually imported account report, comparing source timestamps and principal mapping on connector calls is relevant; this question instead reports that read-only knowledge alone cannot create a service request, so different evidence is needed.
Question 318
Before production promotion, the team needs evidence for this setting: A user asks the agent to read an unrelated executive record. A prior regression caused the following: The model may try a tool outside the user's rights. Which validation is most persuasive?
- Run red-team and representative workflow tests before publishing
- Inject a malicious export instruction into a source document
- Ask for a privileged record using an unprivileged test identity
- Attempt a write from a read-only assistant
Explanation
Correct Answer
C:The historical failure is that the model may try a tool outside the user's rights; for the setting where a user asks the agent to read an unrelated executive record, asking for a privileged record using an unprivileged test identity verifies the dependency where authorization evaluations verify server-side deny behavior.
Incorrect Answers
A:For a hosted agent gains an additional research tool, the alternate test is about whether evaluation suites validate quality and safety before publication; it cannot prove the outcome needed after the reported regression where the model may try a tool outside the user's rights.
B:Injecting a malicious export instruction into a source document verifies an adjacent objective where a prompt injection requests data export to an outside address; it leaves untested the causal path in which authorization evaluations verify server-side deny behavior for this case.
D:This check exercises a different system behavior: distinct permission surfaces enforce least privilege by task; a pass there cannot establish that the model may try a tool outside the user's rights no longer occurs in the required workflow.
Question 319
One enterprise rollout has 2 separately testable requirements. (1) Business use: Refund, credit, reimbursement and adjustment have different accounting impacts. Existing shortcoming: Agent-generated language promises the wrong financial remedy. (2) Business use: Sales users say deal stage differently across regions. Existing shortcoming: An agent updates the wrong pipeline field from an informal phrase. Which 2 success metrics directly satisfy these stated cases? Choose TWO.
- Sales task completion under external API impairment
- Fresh authorized account insights from external systems
- Opportunity changes mapped to the intended business field
- Unauthorized financial commitment language rate
Explanation
Correct Answers
C:The outcome opportunity changes mapped to the intended business field measures whether the workflow where sales users say deal stage differently across regions is improving; the scorecard separately needs unauthorized financial commitment language rate.
D:The outcome unauthorized financial commitment language rate measures whether the workflow where refund, credit, reimbursement and adjustment have different accounting impacts is improving; the scorecard separately needs opportunity changes mapped to the intended business field.
Incorrect Answers
A:The alternative measures sales task completion under external API impairment for the situation where external credit service responds slowly during regional outages; it cannot substitute for performance outcomes in the separate named cases: refund, credit, reimbursement and adjustment have different accounting impacts; sales users say deal stage differently across regions.
B:The alternative measures fresh authorized account insights from external systems for the situation where sellers need current credit exposure held outside CRM; it cannot substitute for performance outcomes in the separate named cases: refund, credit, reimbursement and adjustment have different accounting impacts; sales users say deal stage differently across regions.
Question 320
The pilot needs a durable correction: Logs cannot reconstruct which API mutation followed an answer. This affects the workflow in which a request crosses Copilot Studio, a flow and the CRM backend. Which design choice should the architect approve?
- Connect Foundry agent tracing to Application Insights
- Define end-to-end correlation across agent and business APIs
- Measure tool success, retries and side-effect duplication
- Test Finance-to-Supply-Chain exception handling after partial failure
Explanation
Correct Answer
B:The mechanism that correlation identifiers link the distributed operational path fits the scenario where a request crosses Copilot Studio, a flow and the CRM backend; it addresses the defect that logs cannot reconstruct which API mutation followed an answer.
Incorrect Answers
A:In the case where operations need to correlate prompts, model calls and tools in a Foundry workflow, the alternate design depends on agent trace spans provide component-level execution evidence through connected telemetry; it does not correct the target failure, logs cannot reconstruct which API mutation followed an answer.
C:When agents modify records through external APIs, tool metrics reveal mutation failures and retry consequences; in the stated workflow, however, logs cannot reconstruct which API mutation followed an answer; the relevant architectural behavior is that correlation identifiers link the distributed operational path.
D:The alternative addresses a network fault leaves the two apps inconsistent by a design in which failure-injection tests verify retries and compensation between systems; it does not supply the missing relationship for a request crosses Copilot Studio, a flow and the CRM backend, where logs cannot reconstruct which API mutation followed an answer.