Question 100
The pilot needs a durable correction: A high aggregate score masks critical field errors. This affects the workflow in which a document model performs well overall but fails on tax identifiers. Which design choice should the architect approve?
- Assess robustness under realistic drift and noisy input
- Preserve model-compatible preprocessing and postprocessing contracts
- Preserve reproducible evaluation evidence for major model changes
- Validate custom extraction models by field and document class
Explanation
Correct Answer
D:Without the intended behavior, a high aggregate score masks critical field errors; the proposed architecture establishes the specific design principle that stratified field metrics protect high-impact values for the process where a document model performs well overall but fails on tax identifiers.
Incorrect Answers
A:There is a valid place for the mechanism that stress tests measure sensitivity to expected input variation, namely production documents differ from clean training samples; it does not remove the specific defect that a high aggregate score masks critical field errors from this case.
B:This design would make sense if a retraining release changes the tokenizer and label order, particularly where downstream apps interpret correct predictions as the wrong category; the requested architecture instead needs the dependency that stratified field metrics protect high-impact values because a high aggregate score masks critical field errors.
C:This design would make sense if an updated agent performs differently months after release, particularly where old scores cannot be reproduced because prompts and tests changed; the requested architecture instead needs the dependency that stratified field metrics protect high-impact values because a high aggregate score masks critical field errors.
Question 101
One enterprise rollout has 3 separately testable requirements. (1) Business use: Unused agents retain tokens and active integrations. Known defect: Former owners leave behind accessible service accounts. (2) Business use: Employees install external agents with unknown data pathways. Known defect: An unreviewed plugin exports sensitive tenant information. (3) Business use: Personal read-only assistants and payment agents have different risk. Known defect: Every agent receives the same minimal review. Which 3 diagnostic procedures directly satisfy these stated cases? Choose THREE.
- Compare approval records to agent risk classification
- Compare structured output and errors across candidate cohorts
- Compare inactive agents with still-valid tool grants
- Inspect third-party manifests and requested access scopes
Explanation
Correct Answers
A:Comparing approval records to agent risk classification can reveal why every agent receives the same minimal review when personal read-only assistants and payment agents have different risk; that finding is needed alongside the independent diagnosis for unused agents retain tokens and active integrations; employees install external agents with unknown data pathways.
C:Comparing inactive agents with still-valid tool grants can reveal why former owners leave behind accessible service accounts when unused agents retain tokens and active integrations; that finding is needed alongside the independent diagnosis for employees install external agents with unknown data pathways; personal read-only assistants and payment agents have different risk.
D:Inspecting third-party manifests and requested access scopes can reveal why an unreviewed plugin exports sensitive tenant information when employees install external agents with unknown data pathways; that finding is needed alongside the independent diagnosis for unused agents retain tokens and active integrations; personal read-only assistants and payment agents have different risk.
Incorrect Answers
B:This diagnosis is associated with the different failure where a model upgrade breaks downstream response formatting in a workflow where business applications depend on the agent endpoint continuously; it does not directly solve the listed conditions: former owners leave behind accessible service accounts; an unreviewed plugin exports sensitive tenant information; every agent receives the same minimal review.
Question 102
The architecture board has evidence that engineers blame generation although the backend is slow. The affected situation is: Agents regularly exceed an interactive response-time objective. What change most directly establishes the missing behavior?
- Compare traces before and after orchestration configuration changes
- Constrain high-risk tools with action-level policy checks
- Inspect model-call timing separately from integration latency
- Connect Foundry agent tracing to Application Insights
Explanation
Correct Answer
C:Because per-span latency isolates model work, network waits and tool execution, this approach corrects the reported breakdown where engineers blame generation although the backend is slow; the affected situation is that agents regularly exceed an interactive response-time objective.
Incorrect Answers
A:For another type of architecture, trace comparisons reveal shifts in plan selection and step order is useful; this design must instead support agents regularly exceed an interactive response-time objective despite the problem where engineers blame generation although the backend is slow.
B:The alternative addresses model-level refusal alone may not prevent a tool call by a design in which execution-time destination and permission checks block exfiltration; it does not supply the missing relationship for agents regularly exceed an interactive response-time objective, where engineers blame generation although the backend is slow.
D:For another type of architecture, agent trace spans provide component-level execution evidence through connected telemetry is useful; this design must instead support agents regularly exceed an interactive response-time objective despite the problem where engineers blame generation although the backend is slow.
Question 103
An agent rollout is intended for this situation: Marketing specialists need tailored descriptions from existing briefings. The known problem is that a broad action agent can modify records unnecessarily. Which success measure connects to that specific outcome?
- Drafts meeting structure and source-accuracy criteria
- Unauthorized ancillary actions per task
- Analyst-accepted explanations with verified references
- Ticket updates completed with attributable credentials
Explanation
Correct Answer
A:Drafts meeting structure and source-accuracy criteria; this outcome measure belongs to the workflow where marketing specialists need tailored descriptions from existing briefings; a change in it can help detect whether the defect that a broad action agent can modify records unnecessarily is resolved.
Incorrect Answers
B:The alternative tracks unauthorized ancillary actions per task, arising from the mechanism where limited intent and tool scope prevent unnecessary actions; its movement alone says little about the failure that a broad action agent can modify records unnecessarily.
C:Finance asks why a variance was flagged; its intended measurement is analyst-accepted explanations with verified references; substituting that result here would not capture whether a broad action agent can modify records unnecessarily has been corrected.
D:An agent needs to update a governed ticket status; its intended measurement is ticket updates completed with attributable credentials; substituting that result here would not capture whether a broad action agent can modify records unnecessarily has been corrected.
Question 104
A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Vendors send variable-layout invoices and shipping documents. Known defect: Template parsing omits line items on unfamiliar forms. Which KPI should be selected?
- Grounding recall for critical enterprise questions
- Field-level extraction accuracy on representative documents
- Meaning preservation rate on multilingual service samples
- Correct tool selection and input-binding rate
Explanation
Correct Answer
B:This scorecard needs an outcome related to vendors send variable-layout invoices and shipping documents; field-level extraction accuracy on representative documents. That indicator can be checked against the earlier symptom that template parsing omits line items on unfamiliar forms.
Incorrect Answers
A:For the workflow where stakeholders report hallucinations but cannot isolate search errors, grounding recall for critical enterprise questions is an actionable measure; it does not establish whether template parsing omits line items on unfamiliar forms in the situation under review.
C:The proposed metric concerns support documents arrive in several customer languages, whereas the failure in this case is that template parsing omits line items on unfamiliar forms; the indicator needed here is field-level extraction accuracy on representative documents.
D:The alternative tracks correct tool selection and input-binding rate, arising from the mechanism where specific tool metadata helps the planner choose and parameterize actions; its movement alone says little about the failure that template parsing omits line items on unfamiliar forms.
Question 105
Implementation alternatives must be compared for a workflow where a legacy portal has no integration API for a clerical workflow. The pilot shows this gap: Teams propose browser clicks despite an available stable API. Which alternative should lead the solution?
- Use computer use only where supported APIs are unavailable
- Use agent reasoning for uncertain multi-step analysis with bounded tools
- Design clear provenance and confidence cues in generated interfaces
- Use independent holdout data to select a tuned model
Explanation
Correct Answer
A:The mechanism that uI automation can operate application interfaces but requires stability checks fits the scenario where a legacy portal has no integration API for a clerical workflow; it addresses the defect that teams propose browser clicks despite an available stable API.
Incorrect Answers
B:A simple scripted topic cannot handle legitimate variants; the alternate design can address that situation because reasoning can plan through evidence and tools within enforced boundaries, whereas this question needs a solution to teams propose browser clicks despite an available stable API.
C:This design would make sense if users need to distinguish verified fields from AI suggestions, particularly where staff mistake speculative recommendations for recorded facts; the requested architecture instead needs the dependency that uI automation can operate application interfaces but requires stability checks because teams propose browser clicks despite an available stable API.
D:The technical behavior of this alternative is that separation of training and evaluation yields realistic generalization checks; the user case being assessed is a legacy portal has no integration API for a clerical workflow with a different failure: teams propose browser clicks despite an available stable API.
Question 106
A business stakeholder describes this operating situation: Search spans internal manuals with different owners and permissions. Testing instead shows that a model retrieves a draft manual without any accountable owner. What design approach is most appropriate?
- Define cross-service boundaries and ownership before integration
- Tag documents with ownership, sensitivity, and provenance metadata
- Create a curated semantic layer for business metrics
- Ground conversational answers in SharePoint knowledge with permissions
Explanation
Correct Answer
B:Because metadata enables filtering, review, retention and traceable answers, this approach corrects the reported breakdown where a model retrieves a draft manual without any accountable owner; the affected situation is that search spans internal manuals with different owners and permissions.
Incorrect Answers
A:For copilot, Dataverse, Dynamics and custom APIs share one process, this other approach relies on the fact that contracts and ownership boundaries identify responsibility for each state change; the issue here is a model retrieves a draft manual without any accountable owner; that mechanism does not establish the required metadata enables filtering, review, retention and traceable answers.
C:This design would make sense if sales and finance agents must agree on revenue and margin calculations, particularly where separate assistants report contradictory totals for the same period; the requested architecture instead needs the dependency that metadata enables filtering, review, retention and traceable answers because a model retrieves a draft manual without any accountable owner.
D:For employee policies reside in governed Microsoft 365 documents, this other approach relies on the fact that supported SharePoint grounding retrieves organizational content within access controls; the issue here is a model retrieves a draft manual without any accountable owner; that mechanism does not establish the required metadata enables filtering, review, retention and traceable answers.
Question 107
An agent rollout is intended for this situation: Policy documents contain draft, superseded, and currently effective revisions. The known problem is that the agent cites a superseded rule despite a newer policy existing. Which success measure connects to that specific outcome?
- Successful requests during primary endpoint failures
- Citations referencing in-force policy versions
- Ninety-fifth percentile source-to-answer freshness delay
- Accuracy-adjusted cost per completed business task
Explanation
Correct Answer
B:Citations referencing in-force policy versions; this outcome measure belongs to the workflow where policy documents contain draft, superseded, and currently effective revisions; a change in it can help detect whether the defect that the agent cites a superseded rule despite a newer policy existing is resolved.
Incorrect Answers
A:Although successful requests during primary endpoint failures is observable, the affected process involves policy documents contain draft, superseded, and currently effective revisions; its acceptance outcome is citations referencing in-force policy versions, not the alternative measurement.
C:The premise of this indicator is incremental refresh and source-to-index lag monitoring reduce staleness; the architecture being evaluated instead serves policy documents contain draft, superseded, and currently effective revisions, and its success is measured by citations referencing in-force policy versions.
D:Although accuracy-adjusted cost per completed business task is observable, the affected process involves policy documents contain draft, superseded, and currently effective revisions; its acceptance outcome is citations referencing in-force policy versions, not the alternative measurement.
Question 108
A proposed deployment supports this business case: An assistant responds fluently but cites unrelated support articles. Reviewers also warn that missing retrieval evidence might still generate definitive sensitive advice. Which restriction mitigates that concrete risk?
- Do not allow model confidence prose to override missing evidence
- Fail closed for unsupported sensitive answers
- Redact customer data before making examples available to developers
- Reject cached results that omit original security context
- Refuse cost optimization that breaches business-quality thresholds
Explanation
Correct Answer
B:An assistant responds fluently but cites unrelated support articles; its risk scenario is that missing retrieval evidence might still generate definitive sensitive advice; the safeguard fits because retrieval telemetry distinguishes evidence search from answer generation and the restriction addresses that specific exposure.
Incorrect Answers
A:The alternate guard addresses confident model language might override an explicit evidence requirement for policy answers must be tied to a current source; the present threat, missing retrieval evidence might still generate definitive sensitive advice, needs a different enforcement boundary.
C:The alternate boundary mitigates user transcripts pasted into developer tickets might disclose customer details, which is relevant to users report stale answers after a policy refresh; it leaves the current workflow vulnerable to missing retrieval evidence might still generate definitive sensitive advice.
D:Another process could need this guardrail because shared answer caches might drop source ACLs and leak a private document; the design being assessed supports an assistant responds fluently but cites unrelated support articles and has the explicit exposure that missing retrieval evidence might still generate definitive sensitive advice.
E:The alternate guard addresses lower-cost model routing might undermine critical extraction accuracy for a new model reduces price but also degrades exact extractions; the present threat, missing retrieval evidence might still generate definitive sensitive advice, needs a different enforcement boundary.
Question 109
A quality engineer is writing a regression case after the incident where free-form AI output is written to the wrong Dataverse field. The underlying business scenario is: Inspectors submit a short description and need suggested categorization. What should the test exercise?
- Run representative notes and validate schema and exclusions
- Test mixed intents with missing action parameters
- Test borderline utterances against both descriptions
- Submit missing and malformed values and check validation behavior
Explanation
Correct Answer
D:Submitting missing and malformed values and checking validation behavior challenges the path where an app action validates inputs and structured results before saving for the business setting inspectors submit a short description and need suggested categorization; the specific defect to exclude is that free-form AI output is written to the wrong Dataverse field.
Incorrect Answers
A:The verification concerning an agent must turn service notes into a fixed JSON summary checks prompt actions encapsulate reusable inputs and expected outputs against a general conversational turn produces inconsistent output fields; its result says little about the observed issue where free-form AI output is written to the wrong Dataverse field.
B:In a solution where users request a status check and an authorized follow-up action, testing mixed intents with missing action parameters could catch the defect that a single-topic design cannot combine both operations naturally; the current release gate must challenge the failure where free-form AI output is written to the wrong Dataverse field.
C:In the alternate setting where two topics claim nearly identical scheduling intents, the test exercises distinct descriptions and inputs improve topic selection; even a pass would not show that free-form AI output is written to the wrong Dataverse field has been prevented.
Question 110
The acceptance board asks how to verify a corrected design. The setting is: Candidate processes are attractive but permissions and data quality are weak. Last time, the first-wave pilot requests broad access to unreliable records. Which scenario should the team execute?
- Run a representative extraction holdout at production concurrency
- Add a confidential draft and verify it is excluded for unauthorized users
- Compare prioritized roadmap with a risk-adjusted benefit matrix
- Propose a concession above limit and confirm approval enforcement
Explanation
Correct Answer
C:The first-wave pilot requests broad access to unreliable records; a release test must establish that this no longer occurs; comparing prioritized roadmap with a risk-adjusted benefit matrix measures the specific behavior for which readiness assessment places safe high-value workloads ahead of risky ones.
Incorrect Answers
A:In a solution where high-volume narrow extraction needs predictable latency and cost, running a representative extraction holdout at production concurrency could catch the defect that a general large model exceeds per-document unit economics; the current release gate must challenge the failure where the first-wave pilot requests broad access to unreliable records.
B:For search spans internal manuals with different owners and permissions, the alternate test is about whether metadata enables filtering, review, retention and traceable answers; it cannot prove the outcome needed after the reported regression where the first-wave pilot requests broad access to unreliable records.
D:Proposing a concession above limit and confirming approval enforcement verifies an adjacent objective where service, sales and finance jointly decide high-value concessions; it leaves untested the causal path in which readiness assessment places safe high-value workloads ahead of risky ones for this case.