Question 50
A report draft can be refined against a measurable quality rubric but should not loop indefinitely. Which change most directly satisfies this requirement?
- Classify the request and route to a domain-specific workflow with constrained tools.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Remove the unnecessary debate step and retest quality under the target load.
- Use a bounded generate-critique-revise pattern tied to acceptance criteria.
Explanation
Correct Answer
D:Evaluator feedback supplies specific corrections while a retry ceiling prevents endless polishing. Keep the dependency or branch affecting controlled evaluation-driven revision observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:The design concern is real but different: intent-driven branch selection. It does not impose the specific requirement for controlled evaluation-driven revision; a suitable solution must use a bounded generate-critique-revise pattern tied to acceptance criteria and demonstrate the result.
B:This addresses fan-out review with synchronized approval; however, the problem asks for controlled evaluation-driven revision. Without the safeguard to use a bounded generate-critique-revise pattern tied to acceptance criteria, the original failure mode is still possible.
C:That choice would be responsive to optimization of workflow complexity, but the reported defect is different. It leaves controlled evaluation-driven revision unproved because there is no commitment to use a bounded generate-critique-revise pattern tied to acceptance criteria.
Question 51
A procurement analysis involves searching suppliers, calculating totals and legal review, with each part needing different evidence. Which solution best resolves this operational gap?
- Mark mandatory and optional branches, then degrade with evidence of omitted sources.
- Partition the listings and run checks in parallel with rate-limited workers.
- Split the work by accountable outputs, then integrate with explicit contracts.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
Explanation
Correct Answer
C:Decomposition should follow knowledge boundaries and input dependencies rather than arbitrary paragraph lengths.
Incorrect Answers
A:This recommendation targets explicit partial-result semantics at a different point in the workflow. It leaves work packages with verifiable interfaces outside the design contract and does not require the team to split the work by accountable outputs, then integrate with explicit contracts.
B:This is a valid architectural concern when optimizing rate-limited parallel decomposition. Here it leaves work packages with verifiable interfaces unaddressed, whereas the intended change must split the work by accountable outputs, then integrate with explicit contracts before the case can be considered resolved.
D:By itself, this improves fan-out review with synchronized approval, but the case is not asking the architect to optimize that dimension. The requirement is work packages with verifiable interfaces, supported by an approach to split the work by accountable outputs, then integrate with explicit contracts.
Question 52
Two source documents are independently summarized, but the executive summary must reconcile their inconsistent dates. What design decision would best fit these facts?
- Remove the unnecessary debate step and retest quality under the target load.
- Use sequential steps with validation checkpoints between dependent stages.
- Summarize sources in parallel and perform a final cross-source reconciliation stage.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
Explanation
Correct Answer
C:Parallel upstream work saves time; the merge must still detect contradictions before publishing. Keep the dependency or branch affecting parallel summarization with final consistency checking observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:This shifts attention toward optimization of workflow complexity when the relevant decision is about parallel summarization with final consistency checking. Unless the architecture can summarize sources in parallel and perform a final cross-source reconciliation stage, the failure described in the stem remains unresolved.
B:For a workflow whose primary weakness was dependency-aware task chaining, that proposal could be reasonable. Here the unmet obligation is parallel summarization with final consistency checking, and that requires the system to summarize sources in parallel and perform a final cross-source reconciliation stage.
D:It is aimed at a different outcome, fan-out review with synchronized approval. No evidence is given that it provides parallel summarization with final consistency checking; the needed mechanism is to summarize sources in parallel and perform a final cross-source reconciliation stage at the appropriate boundary.
Question 53
A pipeline repeatedly redoes expensive OCR when only the downstream prompt changes. Which intervention most directly addresses the failure?
- Use sequential steps with validation checkpoints between dependent stages.
- Persist intermediate artifacts and invalidate only affected stages.
- Remove the unnecessary debate step and retest quality under the target load.
- Break the task into typed steps with externally verified computations.
Explanation
Correct Answer
B:Checkpointed outputs allow iterative development and recovery without paying for unaffected computations. Keep the dependency or branch affecting selective workflow recomputation observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:There is a legitimate use case for dependency-aware task chaining, but it does not supply the test or enforcement needed for selective workflow recomputation. The design needs to persist intermediate artifacts and invalidate only affected stages.
C:The alternative could reduce risk associated with optimization of workflow complexity, yet the described behavior requires selective workflow recomputation. That outcome depends on the capability to persist intermediate artifacts and invalidate only affected stages.
D:The proposed measure is aimed at inspectable modular calculation, not selective workflow recomputation. The architecture must instead persist intermediate artifacts and invalidate only affected stages; a separate change is still needed for selective workflow recomputation.
Question 54
A risk-scoring process can produce a useful partial answer when one optional vendor feed times out. What should the solution architect prioritize?
- Partition the listings and run checks in parallel with rate-limited workers.
- Mark mandatory and optional branches, then degrade with evidence of omitted sources.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Cancel dependent notifications with an explicit fail-fast state transition whenever the mandatory gate rejects a case.
Explanation
Correct Answer
B:Failure semantics at each branch prevent optional outages from blocking everything or silently erasing gaps. Keep the dependency or branch affecting explicit partial-result semantics observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:By itself, this improves rate-limited parallel decomposition, but the case is not asking the architect to optimize that dimension. The requirement is explicit partial-result semantics, supported by an approach to mark mandatory and optional branches, then degrade with evidence of omitted sources.
C:The option might improve fan-out review with synchronized approval; that remains insufficient for explicit partial-result semantics. The case is resolved by explicit partial-result semantics, which depends on a design that will mark mandatory and optional branches, then degrade with evidence of omitted sources.
D:The proposed action prioritizes fail-fast cancellation of dependent workflow actions. That can coexist with the right solution, but it cannot replace explicit partial-result semantics: the application must still mark mandatory and optional branches, then degrade with evidence of omitted sources.
Question 55
The team places every operation inside one enormous prompt and cannot inspect intermediate calculations. Which design choice best matches the stated constraint?
- Classify the request and route to a domain-specific workflow with constrained tools.
- Persist intermediate artifacts and invalidate only affected stages.
- Break the task into typed steps with externally verified computations.
- Use sequential steps with validation checkpoints between dependent stages.
Explanation
Correct Answer
C:Observable components allow deterministic calculations to be tested separately from language generation. Keep the dependency or branch affecting inspectable modular calculation observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:It is aimed at a different outcome, intent-driven branch selection. No evidence is given that it provides inspectable modular calculation; the needed mechanism is to break the task into typed steps with externally verified computations at the appropriate boundary.
B:It emphasizes selective workflow recomputation without closing the stated gap in inspectable modular calculation. By contrast, the remedy to break the task into typed steps with externally verified computations acts on the condition that makes this scenario fail.
D:Improving dependency-aware task chaining does not imply inspectable modular calculation. The selected design must break the task into typed steps with externally verified computations, whereas this choice omits that direct protection or verification step.
Question 56
A service workflow contains an unhelpful debate agent that rarely changes classification. It also sends outbound notices after a mandatory compliance gate fails, and its interactive requests wait for slow exception investigations that could run asynchronously. Which THREE workflow changes fit these separate problems? Choose THREE.
- Persist intermediate artifacts and invalidate only affected stages.
- Cancel dependent notifications with an explicit fail-fast state transition whenever the mandatory gate rejects a case.
- Remove the unnecessary debate step and retest quality under the target load.
- Split interactive work from asynchronous exception investigation using separate task-class deadlines and result delivery.
- Summarize sources in parallel and perform a final cross-source reconciliation stage.
Explanation
Correct Answers
B:A rejected mandatory checkpoint must stop its downstream effects rather than merely add an error note to the trace. An explicit terminal state prevents later jobs from treating an invalid case as publishable.
C:The debate stage consumes additional inference while rarely altering the result. Removing it after a quality-regression check reduces unnecessary cost and latency; the independent fail-fast and asynchronous requirements are enforced by separate state and scheduling controls.
D:The common path should not inherit the delay budget of infrequent deep investigations. A queued exception path provides a completion mechanism without keeping ordinary requests blocked on slow analysis.
Incorrect Answers
A:The proposed measure is aimed at selective workflow recomputation, not optimization of workflow complexity and fail-fast cancellation of dependent workflow actions and asynchronous exception processing without blocking fast paths. The architecture must instead split interactive work from asynchronous exception investigation using separate task-class deadlines and result delivery; a separate change is still needed for optimization of workflow complexity and fail-fast cancellation of dependent workflow actions and asynchronous exception processing without blocking fast paths.
E:This shifts attention toward parallel summarization with final consistency checking when the relevant decision is about optimization of workflow complexity and fail-fast cancellation of dependent workflow actions and asynchronous exception processing without blocking fast paths. Unless the architecture can cancel dependent notifications with an explicit fail-fast state transition whenever the mandatory gate rejects a case, the failure described in the stem remains unresolved.
Question 57
A policy-checking pipeline requires independent privacy and security sign-offs before release. Which approach should be recommended?
- Mark mandatory and optional branches, then degrade with evidence of omitted sources.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Partition the listings and run checks in parallel with rate-limited workers.
- Persist intermediate artifacts and invalidate only affected stages.
Explanation
Correct Answer
B:Parallel review can reduce latency while a synchronized gate preserves dual-control requirements. Keep the dependency or branch affecting fan-out review with synchronized approval observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:The proposed action prioritizes explicit partial-result semantics. That can coexist with the right solution, but it cannot replace fan-out review with synchronized approval: the application must still fan out to both reviewers, then gate publication on their combined approved outcomes.
C:This addresses rate-limited parallel decomposition; however, the problem asks for fan-out review with synchronized approval. Without the safeguard to fan out to both reviewers, then gate publication on their combined approved outcomes, the original failure mode is still possible.
D:That choice would be responsive to selective workflow recomputation, but the reported defect is different. It leaves fan-out review with synchronized approval unproved because there is no commitment to fan out to both reviewers, then gate publication on their combined approved outcomes.
Question 58
A contact center reduces drafting time but doubles supervisor review work, worsening total handle time. What is the best production design response?
- Model token, retrieval, tooling and peak concurrency costs with conservative headroom.
- Segment the workflow by complexity and send low-confidence exceptions to human handling.
- Negotiate measurable coverage, citation quality, abstention and review controls instead of impossible guarantees.
- Measure end-to-end cost and quality rather than model inference duration alone.
Explanation
Correct Answer
D:Net operational value includes human correction, escalation and rework, not just faster first drafts. Evaluate the whole operating process for full-workflow return on investment, including human review and failures, rather than only the API-level benchmark.
Incorrect Answers
A:Improving sustainable cost at peak usage does not imply full-workflow return on investment. The selected design must measure end-to-end cost and quality rather than model inference duration alone, whereas this choice omits that direct protection or verification step.
B:This recommendation targets selective automation by risk at a different point in the workflow. It leaves full-workflow return on investment outside the design contract and does not require the team to measure end-to-end cost and quality rather than model inference duration alone.
C:This is a valid architectural concern when optimizing realistic quality commitments. Here it leaves full-workflow return on investment unaddressed, whereas the intended change must measure end-to-end cost and quality rather than model inference duration alone before the case can be considered resolved.
Question 59
A warehouse assistant meets quality targets but exceeds its monthly operating budget during seasonal peaks. Which change most directly satisfies this requirement?
- Weight evaluation and safeguards by consequence, not by frequency alone.
- Benchmark risk-adjusted total cost including retries and review effort before switching tiers.
- Measure end-to-end cost and quality rather than model inference duration alone.
- Model token, retrieval, tooling and peak concurrency costs with conservative headroom.
Explanation
Correct Answer
D:Architecture selection must remain viable at expected volume, not only at pilot traffic. Evaluate the whole operating process for sustainable cost at peak usage, including human review and failures, rather than only the API-level benchmark.
Incorrect Answers
A:The option belongs to the broader work on risk-adjusted performance priorities, but it cannot by itself deliver sustainable cost at peak usage. A decision on this case needs the team to model token, retrieval, tooling and peak concurrency costs with conservative headroom and verify that outcome.
B:This shifts attention toward quality-adjusted model economics when the relevant decision is about sustainable cost at peak usage. Unless the architecture can model token, retrieval, tooling and peak concurrency costs with conservative headroom, the failure described in the stem remains unresolved.
C:For a workflow whose primary weakness was full-workflow return on investment, that proposal could be reasonable. Here the unmet obligation is sustainable cost at peak usage, and that requires the system to model token, retrieval, tooling and peak concurrency costs with conservative headroom.
Question 60
An executive requests 100% factual correctness for an open-ended knowledge assistant. Which solution best resolves this operational gap?
- Benchmark risk-adjusted total cost including retries and review effort before switching tiers.
- Measure end-to-end cost and quality rather than model inference duration alone.
- Negotiate measurable coverage, citation quality, abstention and review controls instead of impossible guarantees.
- Model token, retrieval, tooling and peak concurrency costs with conservative headroom.
Explanation
Correct Answer
C:Operational commitments can describe tested error rates and escalation, not universal truth for every possible query.
Incorrect Answers
A:That choice would be responsive to quality-adjusted model economics, but the reported defect is different. It leaves realistic quality commitments unproved because there is no commitment to negotiate measurable coverage, citation quality, abstention and review controls instead of impossible guarantees.
B:There is a legitimate use case for full-workflow return on investment, but it does not supply the test or enforcement needed for realistic quality commitments. The design needs to negotiate measurable coverage, citation quality, abstention and review controls instead of impossible guarantees.
D:The alternative could reduce risk associated with sustainable cost at peak usage, yet the described behavior requires realistic quality commitments. That outcome depends on the capability to negotiate measurable coverage, citation quality, abstention and review controls instead of impossible guarantees.