Question 60
A helpdesk sponsor asks for an AI assistant that is 'accurate and fast.' What should engineering define before choosing a model?
- Specify supported languages and measure task quality for each representative locale.
- Approximate document volume, query rate, indexing load, and latency budgets.
- Convert the request into measurable quality, latency, throughput, and risk targets.
- Evaluate source-supported claims and rejection of unsupported questions on a held-out set.
Explanation
Correct Answer
C:Acceptance criteria turn adjectives into testable constraints; model selection without task metrics can optimize the wrong property.
Incorrect Answers
A:Language performance varies by input distribution and task type, so a single English evaluation cannot establish fitness; it follows multilingual quality targets constraints but leaves measurable product requirements unspecified.
B:Retrieval storage and preprocessing costs differ from model inference; capacity planning needs both workloads; the effect concerns retrieval infrastructure sizing; the success condition is measurable product requirements.
D:Grounding is a product behavior that needs verification against retrieval evidence, not a vague instruction to avoid hallucinations; this may help grounded answer acceptance but cannot by itself deliver measurable product requirements.
Question 61
A claims team handles regulated customer records and wants external-model summarization. What requirement must be identified before the API architecture is approved?
- Specify degraded-mode behavior and safe dependency-failure responses.
- Document allowed data flows, retention obligations, and residency constraints.
- Define authorization roles and approval thresholds for each mutating action.
- As an alternative, convert the request into measurable quality, latency, throughput, and risk targets.
Explanation
Correct Answer
B:Information governance determines which data can cross a service boundary and which controls the design must provide.
Incorrect Answers
A:Graceful degradation lets the application preserve read-only functionality while refusing actions that require unavailable authoritative systems; a remedy for dependency failure policy is not a substitute for regulated data boundaries.
C:Tool design depends on who can commit side effects; missing approval requirements cannot be repaired by prompt wording; the consequence is improved action authorization requirements, not verified regulated data boundaries.
D:Acceptance criteria turn adjectives into testable constraints; model selection without task metrics can optimize the wrong property; it improves measurable product requirements only, leaving regulated data boundaries unproven.
Question 62
A user-facing chat product has a strict two-second response experience but variable generation length. Which requirement should the interface specification prioritize?
- Establish supported languages and measure task quality for each representative locale.
- Define perceived latency and progressive-response behavior, including streaming.
- As an alternative, establish acceptable accuracy, latency, and cost per task category.
- Specify which decisions, tool calls, versions, and approvals must be traceable.
- Specify a shared application service contract and client-specific interaction needs.
Explanation
Correct Answer
B:Time to first useful output may matter more than complete-generation time for an interactive interface; the SLA should reflect that.
Incorrect Answers
A:Language performance varies by input distribution and task type, so a single English evaluation cannot establish fitness; for conversational latency requirements, a control centered on multilingual quality targets is not the appropriate substitute.
C:A single maximum-capability configuration wastes resources when simpler tasks meet targets with lower-cost or faster models; this targets task-specific economic constraints, while the system must demonstrate conversational latency requirements.
D:Audit logs are useful only when they capture the events needed to reconstruct actions and establish accountable control; this implements multi-criterion integration rather than the demanded conversational latency requirements.
E:A stable backend API can centralize model rules and security, while each client has different presentation and latency expectations; the scenario's controlling requirement is conversational latency requirements, not the related multichannel integration boundaries concern.
Question 63
A nightly document-classification job can finish any time before morning. What should the team clarify to choose its processing mode?
- Establish a shared application service contract and client-specific interaction needs.
- Specify perceived latency and progressive-response behavior, including streaming.
- Separate asynchronous completion deadlines from interactive response requirements.
- Project document volume, query rate, indexing load, and latency budgets.
Explanation
Correct Answer
C:Deferred workloads can be queued and batched for efficiency, while live requests need immediate request-response behavior.
Incorrect Answers
A:A stable backend API can centralize model rules and security, while each client has different presentation and latency expectations; for this request, deferred completion SLA is the missing piece, not multichannel integration boundaries.
B:Time to first useful output may matter more than complete-generation time for an interactive interface; the SLA should reflect that; it changes conversational latency requirements; it does not prove deferred completion SLA in this case.
D:Retrieval storage and preprocessing costs differ from model inference; capacity planning needs both workloads; where deferred completion SLA matters, a control for retrieval infrastructure sizing leaves that test unmet.
Question 64
A business requests a tool-capable assistant for order changes but cannot explain who may approve refunds. What should be resolved first?
- Define which decisions, tool calls, versions, and approvals must be traceable.
- Specify authorization roles and approval thresholds for each mutating action.
- As an alternative, fix representative tasks, success criteria, and operational budget for both approaches.
- Establish perceived latency and progressive-response behavior, including streaming.
Explanation
Correct Answer
B:Tool design depends on who can commit side effects; missing approval requirements cannot be repaired by prompt wording.
Incorrect Answers
A:Audit logs are useful only when they capture the events needed to reconstruct actions and establish accountable control; a remedy for multi-criterion integration is not a substitute for action authorization requirements.
C:Comparative evaluation is only meaningful when task difficulty, measured outcomes, and cost constraints are held consistent; because architecture evaluation baseline is its concern, action authorization requirements remains outside its effect here.
D:Time to first useful output may matter more than complete-generation time for an interactive interface; the SLA should reflect that; for action authorization requirements, a control centered on conversational latency requirements is not the appropriate substitute.
Question 65
The same Claude feature will serve mobile, web, and internal batch clients. What should the requirements capture?
- For this workflow, convert the request into measurable quality, latency, throughput, and risk targets.
- Establish degraded-mode behavior and safe dependency-failure responses.
- Write down allowed data flows, retention obligations, and residency constraints.
- Define a shared application service contract and client-specific interaction needs.
Explanation
Correct Answer
D:A stable backend API can centralize model rules and security, while each client has different presentation and latency expectations.
Incorrect Answers
A:Acceptance criteria turn adjectives into testable constraints; model selection without task metrics can optimize the wrong property; a remedy for measurable product requirements is not a substitute for multichannel integration boundaries.
B:Graceful degradation lets the application preserve read-only functionality while refusing actions that require unavailable authoritative systems; the consequence is improved dependency failure policy, not verified multichannel integration boundaries.
C:Information governance determines which data can cross a service boundary and which controls the design must provide; changing regulated data boundaries does not establish multichannel integration boundaries in this application.
Question 66
A technical leader proposes the strongest available model for every request. Which missing requirement would make that an unsound default?
- At the application layer, convert the request into measurable quality, latency, throughput, and risk targets.
- Isolate extraction and recommendation from irreversible payment authority.
- Establish acceptable accuracy, latency, and cost per task category.
- Isolate asynchronous completion deadlines from interactive response requirements.
Explanation
Correct Answer
C:A single maximum-capability configuration wastes resources when simpler tasks meet targets with lower-cost or faster models.
Incorrect Answers
A:Acceptance criteria turn adjectives into testable constraints; model selection without task metrics can optimize the wrong property; when the goal is task-specific economic constraints, optimizing measurable product requirements misses that operational boundary.
B:Payment involves external side effects and fraud exposure; deterministic approvals and authorization controls belong in the requirements; this may help financially consequential automation but cannot by itself deliver task-specific economic constraints.
D:Deferred workloads can be queued and batched for efficiency, while live requests need immediate request-response behavior; it changes deferred completion SLA; it does not prove task-specific economic constraints in this case.
Question 67
A knowledge assistant must answer only from approved internal manuals. Which acceptance criterion tests that requirement?
- Document a shared application service contract and client-specific interaction needs.
- Document degraded-mode behavior and safe dependency-failure responses.
- Partition asynchronous completion deadlines from interactive response requirements.
- Record allowed data flows, retention obligations, and residency constraints.
- Score source-supported claims and rejection of unsupported questions on a held-out set.
- For this workflow, establish acceptable accuracy, latency, and cost per task category.
Explanation
Correct Answer
E:Grounding is a product behavior that needs verification against retrieval evidence, not a vague instruction to avoid hallucinations.
Incorrect Answers
A:A stable backend API can centralize model rules and security, while each client has different presentation and latency expectations; even if multichannel integration boundaries succeeds, grounded answer acceptance still needs separate verification.
B:Graceful degradation lets the application preserve read-only functionality while refusing actions that require unavailable authoritative systems; because dependency failure policy is its concern, grounded answer acceptance remains outside its effect here.
C:Deferred workloads can be queued and batched for efficiency, while live requests need immediate request-response behavior; for grounded answer acceptance, a control centered on deferred completion SLA is not the appropriate substitute.
D:Information governance determines which data can cross a service boundary and which controls the design must provide; this targets regulated data boundaries, while the system must demonstrate grounded answer acceptance.
F:A single maximum-capability configuration wastes resources when simpler tasks meet targets with lower-cost or faster models; the consequence is improved task-specific economic constraints, not verified grounded answer acceptance.
Question 68
A new service must process multilingual customer complaints. Which requirement needs empirical clarification?
- Grade source-supported claims and rejection of unsupported questions on a held-out set.
- For this workflow, fix representative tasks, success criteria, and operational budget for both approaches.
- Partition extraction and recommendation from irreversible payment authority.
- Define supported languages and measure task quality for each representative locale.
Explanation
Correct Answer
D:Language performance varies by input distribution and task type, so a single English evaluation cannot establish fitness.
Incorrect Answers
A:Grounding is a product behavior that needs verification against retrieval evidence, not a vague instruction to avoid hallucinations; the scenario's controlling requirement is multilingual quality targets, not the related grounded answer acceptance concern.
B:Comparative evaluation is only meaningful when task difficulty, measured outcomes, and cost constraints are held consistent; the multilingual quality targets criterion differs from architecture evaluation baseline; the gap persists.
C:Payment involves external side effects and fraud exposure; deterministic approvals and authorization controls belong in the requirements; it improves financially consequential automation only, leaving multilingual quality targets unproven.
Question 69
A product owner demands fully automatic invoice payment based on emails. What risk assessment should precede implementation?
- Separate extraction and recommendation from irreversible payment authority.
- Document perceived latency and progressive-response behavior, including streaming.
- At the application layer, establish acceptable accuracy, latency, and cost per task category.
- Calculate document volume, query rate, indexing load, and latency budgets.
Explanation
Correct Answer
A:Payment involves external side effects and fraud exposure; deterministic approvals and authorization controls belong in the requirements.
Incorrect Answers
B:Time to first useful output may matter more than complete-generation time for an interactive interface; the SLA should reflect that; this implements conversational latency requirements rather than the demanded financially consequential automation.
C:A single maximum-capability configuration wastes resources when simpler tasks meet targets with lower-cost or faster models; because task-specific economic constraints is its concern, financially consequential automation remains outside its effect here.
D:Retrieval storage and preprocessing costs differ from model inference; capacity planning needs both workloads; this may help retrieval infrastructure sizing but cannot by itself deliver financially consequential automation.
Question 70
A retrieval product will ingest thousands of PDFs. What infrastructure input must be sized before choosing the request topology?
- Specify allowed data flows, retention obligations, and residency constraints.
- In this design, convert the request into measurable quality, latency, throughput, and risk targets.
- Document supported languages and measure task quality for each representative locale.
- Estimate document volume, query rate, indexing load, and latency budgets.
- Set out perceived latency and progressive-response behavior, including streaming.
Explanation
Correct Answer
D:Retrieval storage and preprocessing costs differ from model inference; capacity planning needs both workloads.
Incorrect Answers
A:Information governance determines which data can cross a service boundary and which controls the design must provide; the scenario's controlling requirement is retrieval infrastructure sizing, not the related regulated data boundaries concern.
B:Acceptance criteria turn adjectives into testable constraints; model selection without task metrics can optimize the wrong property; the effect concerns measurable product requirements; the success condition is retrieval infrastructure sizing.
C:Language performance varies by input distribution and task type, so a single English evaluation cannot establish fitness; when the goal is retrieval infrastructure sizing, optimizing multilingual quality targets misses that operational boundary.
E:Time to first useful output may matter more than complete-generation time for an interactive interface; the SLA should reflect that; this targets conversational latency requirements, while the system must demonstrate retrieval infrastructure sizing.