Question 850
Before introducing a fix, the team must explain why a prompt update changes decisions for all users at once. The affected setting is: New behavior must be evaluated without exposing the full workforce. What should be inspected?
- Inspect write conflicts and source-of-truth definitions
- Inspect action failures for bad types and entity mismatches
- Audit rollout settings and incident timelines for uncontrolled changes
- Check whether the required API can be surfaced through supported actions
Explanation
Correct Answer
C:The investigation directly tests the causal connection: controlled release gates restrict blast radius and support rollback; it collects evidence for a workflow where new behavior must be evaluated without exposing the full workforce but the outcome is instead that a prompt update changes decisions for all users at once.
Incorrect Answers
A:The evidence this method produces applies to a workflow where two applications try to overwrite the same customer credit field; it cannot by itself establish why a prompt update changes decisions for all users at once in the current scenario.
B:The proposed probe concerns the model submits malformed identifiers to an operational API, where typed inputs and server validation reject malformed or out-of-scope calls; this incident instead requires evidence about controlled release gates restrict blast radius and support rollback and the failure that a prompt update changes decisions for all users at once.
D:Checking whether the required api can be surfaced through supported actions explores a separate connection, where supported extensibility adds an action while retaining the host's orchestration; the diagnosed requirement is new behavior must be evaluated without exposing the full workforce, and the actual defect is a prompt update changes decisions for all users at once.
Question 851
A staging test has to detect the specific defect that it substitutes a prior customer's order in a later turn. In the workflow where agent must remember order context while asking for missing details, which exercise verifies the corrective behavior?
- Send concurrent equivalent events and inspect resulting writes
- Reproduce a sanitized failure and confirm the new regression catches it
- Alternate ambiguous and corrected identifiers within one session
- Execute both approved and rejected concession scenarios
Explanation
Correct Answer
C:For the required workflow where agent must remember order context while asking for missing details, the test needs to reproduce or prevent the failure where it substitutes a prior customer's order in a later turn; alternating ambiguous and corrected identifiers within one session exercises the path where multi-turn tests assess state tracking, entity separation and follow-up questions.
Incorrect Answers
A:In the alternate setting where several events refer to one business order at once, the test exercises concurrency evaluation tests idempotency and serialization; even a pass would not show that it substitutes a prior customer's order in a later turn has been prevented.
B:In a solution where support tickets reveal an unusual multistep conversation failure, reproducing a sanitized failure and confirming the new regression catches it could catch the defect that the fixed scenario is not added to future test suites; the current release gate must challenge the failure where it substitutes a prior customer's order in a later turn.
D:Executing both approved and rejected concession scenarios verifies an adjacent objective where service recommends a price adjustment requiring finance approval; it leaves untested the causal path in which multi-turn tests assess state tracking, entity separation and follow-up questions for this case.
Question 852
The fault affects an AI-powered workflow in which a backend agent must call an approved Azure service without embedded secrets. Specifically, long-lived API keys leak from configuration and diagnostics. Which diagnostic investigation best distinguishes the relevant failure path?
- Compare ticket assignment confusion matrices across versions
- Compare action privileges before and after package import
- Review endpoint access logs and allowed network paths
- Audit token use and secret inventory for agent services
Explanation
Correct Answer
D:For the case where a backend agent must call an approved Azure service without embedded secrets, the diagnostic path must explain the failure that long-lived API keys leak from configuration and diagnostics; auditing token use and secret inventory for agent services targets the mechanism by which managed identity provides credential-less workload authentication where supported.
Incorrect Answers
A:That probe evaluates controlled service evaluations validate routing before rollout in a workflow where a support agent changes how it classifies priority intents; the reported failure concerns the path for which managed identity provides credential-less workload authentication where supported instead.
B:That probe evaluates authorization regression gates verify posting authority remains constrained in a workflow where agents suggest payment adjustments and inventory holds; the reported failure concerns the path for which managed identity provides credential-less workload authentication where supported instead.
C:For the failure that untrusted clients submit requests and exfiltrate outputs, reviewing endpoint access logs and allowed network paths is relevant; this question instead reports that long-lived API keys leak from configuration and diagnostics, so different evidence is needed.
Question 853
Support engineers report this failure: A costly custom runtime replicates a mature standard capability. The business context is: Users need ordinary meeting summaries and follow-up drafts. What evidence should be collected to isolate the cause?
- Compare baseline model errors by language and domain
- Inspect task envelopes and agent capability declarations
- Compare required outcomes with the prebuilt agent's supported boundaries
- Trace latency and authorization across the managed/custom boundary
Explanation
Correct Answer
C:The reported fault is that a costly custom runtime replicates a mature standard capability in a system where users need ordinary meeting summaries and follow-up drafts; evidence from comparing required outcomes with the prebuilt agent's supported boundaries can show whether a prebuilt Microsoft 365 capability minimizes custom maintenance for commodity tasks.
Incorrect Answers
A:The proposed inspection is associated with a unique domain vocabulary produces persistent measured errors, where a measured capability gap, appropriate dataset and MLOps plan justify customization; it would not expose the failure mode that a costly custom runtime replicates a mature standard capability with the required specificity.
B:The evidence this method produces applies to a workflow where independently hosted agents must exchange tasks and results; it cannot by itself establish why a costly custom runtime replicates a mature standard capability in the current scenario.
D:Tracing latency and authorization across the managed/custom boundary explores a separate connection, where a managed host can invoke an independently governed domain API; the diagnosed requirement is users need ordinary meeting summaries and follow-up drafts, and the actual defect is a costly custom runtime replicates a mature standard capability.
Question 854
In a business situation where a sensitive model endpoint is reachable using a broadly shared key, a previous build showed this defect: Untrusted clients submit requests and exfiltrate outputs. Which reproducible test belongs in the release gate?
- Shift limited traffic to a candidate and monitor regressions
- Attempt inference from an unauthorized identity or network
- Make identical queries as permitted and denied users
- Try a forbidden refund despite a permissive model instruction
- Interrupt processing after the first app commits
- Query the same protected record from authorized and denied identities
Explanation
Correct Answer
B:Attempting inference from an unauthorized identity or network is aligned with the failure that untrusted clients submit requests and exfiltrate outputs; its causal relevance is that identity-based access and supported network isolation reduce exposure for the situation where a sensitive model endpoint is reachable using a broadly shared key.
Incorrect Answers
A:The tested behavior is that canary traffic and versioned rollback reduce release risk; the necessary evidence here concerns identity-based access and supported network isolation reduce exposure and the previously observed untrusted clients submit requests and exfiltrate outputs.
C:Although the exercise might catch one user's cached confidential answer appears for another user, it would not directly validate the mechanism that identity-based access and supported network isolation reduce exposure in the target implementation.
D:In the alternate setting where prompt wording persuades the agent to submit a privileged operation, the test exercises server-side access checks enforce real actor entitlements; even a pass would not show that untrusted clients submit requests and exfiltrate outputs has been prevented.
E:A pass on that verification would speak to a blocked invoice triggers a hold on related procurement actions; it would not show that identity-based access and supported network isolation reduce exposure, which is needed to avoid the observed defect untrusted clients submit requests and exfiltrate outputs.
F:In the alternate setting where agent looks up records on behalf of employees with different rights, the test exercises delegated authorization evaluates each caller's effective permissions; even a pass would not show that untrusted clients submit requests and exfiltrate outputs has been prevented.
Question 855
The architect wants evidence before altering the deployment. In a workflow where data scientists cannot explain why a newer classifier behaves differently, models move to production without an audit of training inputs. What should the team examine first?
- Inspect source identifiers and ingestion timestamps in index records
- Compare errors by scan quality and source drift
- Review the signed acceptance matrix and results
- Inspect artifacts for training data hashes and evaluation reports
- Match release manifest to transformation and model checksums
Explanation
Correct Answer
D:For the situation where data scientists cannot explain why a newer classifier behaves differently, the defect is that models move to production without an audit of training inputs; inspecting artifacts for training data hashes and evaluation reports helps verify the cause-and-effect path: a model registry ties versions to lineage, evaluations and approval.
Incorrect Answers
A:The proposed probe concerns agents begin citing an obsolete article after an ingestion job, where content lineage records update, transformation and index revision; this incident instead requires evidence about a model registry ties versions to lineage, evaluations and approval and the failure that models move to production without an audit of training inputs.
B:For the failure that scanned forms suddenly produce inaccurate predictions, comparing errors by scan quality and source drift is relevant; this question instead reports that models move to production without an audit of training inputs, so different evidence is needed.
C:In the process where teams promote a custom model because it feels better, the diagnostic procedure could explain candidate improves one metric while violating another; this particular incident instead turns on whether a model registry ties versions to lineage, evaluations and approval and why models move to production without an audit of training inputs.
E:That probe evaluates versioned data transformations and model artifacts enable consistent rollback in a workflow where a preprocessing release changes category labels incorrectly; the reported failure concerns the path for which a model registry ties versions to lineage, evaluations and approval instead.
Question 856
A staging test has to detect the specific defect that overall performance masks unequal escalation rates. In the workflow where an AI triage model under-serves one customer language, which exercise verifies the corrective behavior?
- Replay a stable request set on both prompt versions
- Test equivalent requests in represented languages
- Try to send high-impact notices without approval in staging
- Roll out to one region and check segmented resolution metrics
Explanation
Correct Answer
B:The historical failure is that overall performance masks unequal escalation rates; for the setting where an AI triage model under-serves one customer language, testing equivalent requests in represented languages verifies the dependency where segment-aware evaluations surface fairness risks.
Incorrect Answers
A:In the alternate setting where model bills increase although request volume is stable, the test exercises token telemetry isolates prompt, retrieval and generation cost changes; even a pass would not show that overall performance masks unequal escalation rates has been prevented.
C:This check exercises a different system behavior: business approval checks remain independent of agent version; a pass there cannot establish that overall performance masks unequal escalation rates no longer occurs in the required workflow.
D:In a solution where a conversational change affects only one support geography, rolling out to one region and checking segmented resolution metrics could catch the defect that global averages hide a regional escalation spike; the current release gate must challenge the failure where overall performance masks unequal escalation rates.
Question 857
A business stakeholder describes this operating situation: The agent may refund large amounts or deactivate accounts. Testing instead shows that the conversation alone authorizes a permanent business change. What design approach is most appropriate?
- Use staged rollout with feature flags and pilot cohorts
- Build a bespoke agent when differentiated orchestration is essential
- Implement an abstain-or-escalate response policy for uncertain outputs
- Require approval before irreversible agent actions
Explanation
Correct Answer
D:The defect is that the conversation alone authorizes a permanent business change; the expected business case is that the agent may refund large amounts or deactivate accounts; the design fits because confirmation and human approval separate advice from execution authority.
Incorrect Answers
A:The technical behavior of this alternative is that controlled release gates restrict blast radius and support rollback; the user case being assessed is the agent may refund large amounts or deactivate accounts with a different failure: the conversation alone authorizes a permanent business change.
B:When competitive requirements depend on unique multi-model control, custom engineering trades more control for more operating responsibility; in the stated workflow, however, the conversation alone authorizes a permanent business change; the relevant architectural behavior is that confirmation and human approval separate advice from execution authority.
C:Confidence and evidence checks route insufficiently supported cases to humans; its main purpose is the system cannot always find evidence for regulated guidance, not correction of the reported defect where the conversation alone authorizes a permanent business change in the target process.
Question 858
Release reviewers will not accept a generic success message. The business scenario is that agents must distinguish an incident, complaint, inquiry and entitlement, and the historical failure is: Copilot misclassifies a complaint as an ordinary information request. Which test is appropriately targeted?
- Delay the external API and verify useful partial handling
- Query synonyms and near-neighbor phrases and check intended classification
- Test duplicate legal names with different external IDs
- Change a permitted credit field and verify updated authorized retrieval
Explanation
Correct Answer
B:The scenario is that agents must distinguish an incident, complaint, inquiry and entitlement; the defect to rule out is that Copilot misclassifies a complaint as an ordinary information request; testing via querying synonyms and near-neighbor phrases and checking intended classification challenges the design dependency where business-term definitions align user vocabulary with governed service entities.
Incorrect Answers
A:The suggested test serves external credit service responds slowly during regional outages, in which a sales workflow hangs awaiting optional enrichment; the present scenario instead concerns the reported problem where Copilot misclassifies a complaint as an ordinary information request.
C:A pass on that verification would speak to two systems share company names but not primary keys; it would not show that business-term definitions align user vocabulary with governed service entities, which is needed to avoid the observed defect Copilot misclassifies a complaint as an ordinary information request.
D:The verification concerning sellers need current credit exposure held outside CRM checks a supported connector can retrieve external information with explicit identity mapping against Copilot relies on last quarter's manually imported account report; its result says little about the observed issue where Copilot misclassifies a complaint as an ordinary information request.
Question 859
Release reviewers will not accept a generic success message. The business scenario is that customers ask for current delivery or payment status, and the historical failure is: A cached knowledge answer is several hours old. Which test is appropriately targeted?
- Change a status and verify immediate answer consistency
- Test mixed-domain requests and validate specialist selection
- Call a read-only tool with valid and invalid identity contexts
- Disconnect the source and verify a transparent failure response
Explanation
Correct Answer
A:A cached knowledge answer is several hours old; a release test must establish that this no longer occurs; changing a status and verifying immediate answer consistency measures the specific behavior for which authorized actions query the system of record at request time.
Incorrect Answers
B:The tested behavior is that a coordinating agent selects bounded specialists and merges their outputs; the necessary evidence here concerns authorized actions query the system of record at request time and the previously observed a cached knowledge answer is several hours old.
C:The tested behavior is that MCP presents supported tools and resources through a common integration contract; the necessary evidence here concerns authorized actions query the system of record at request time and the previously observed a cached knowledge answer is several hours old.
D:This check exercises a different system behavior: health checks and explicit no-evidence responses handle missing context safely; a pass there cannot establish that a cached knowledge answer is several hours old no longer occurs in the required workflow.
Question 860
The architecture board has evidence that no log identifies which source version was indexed. The affected situation is: Agents begin citing an obsolete article after an ingestion job. What change most directly establishes the missing behavior?
- Preserve reproducible evaluation evidence for major model changes
- Restrict who can supply tuning data and approve model updates
- Track grounding-source updates and index rebuild provenance
- Retain rollback snapshots for models and data transformations
Explanation
Correct Answer
C:The mechanism that content lineage records update, transformation and index revision fits the scenario where agents begin citing an obsolete article after an ingestion job; it addresses the defect that no log identifies which source version was indexed.
Incorrect Answers
A:When an updated agent performs differently months after release, versioned evaluation sets, model IDs and thresholds document release claims; in the stated workflow, however, no log identifies which source version was indexed; the relevant architectural behavior is that content lineage records update, transformation and index revision.
B:For another type of architecture, separated data-curation and release permissions preserve integrity is useful; this design must instead support agents begin citing an obsolete article after an ingestion job despite the problem where no log identifies which source version was indexed.
D:The alternative addresses restoring code alone does not restore previous predictions by a design in which versioned data transformations and model artifacts enable consistent rollback; it does not supply the missing relationship for agents begin citing an obsolete article after an ingestion job, where no log identifies which source version was indexed.