Question 330
A legal stakeholder worries that an autonomous agent may send unapproved letters to customers. Which change most directly satisfies this requirement?
- Compare update frequency, evidence traceability and data ownership in a decision matrix.
- Show the distinction between behavioral instructions and gateway-enforced transaction permissions.
- Describe region-specific decision constraints and a common control baseline.
- Draw a workflow showing draft, evidence validation, human approval and outbound-send controls.
Explanation
Correct Answer
D:A clear control path communicates responsibility more effectively than generic assurances about safe AI. Show evidence for visible accountability for outbound action alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
A:The control would be justified if fit-based retrieval-versus-customization reasoning were the limiting factor. In this scenario the decisive target is visible accountability for outbound action; success requires an architecture that will draw a workflow showing draft, evidence validation, human approval and outbound-send controls.
B:The option belongs to the broader work on clear explanation of enforceable safeguards, but it cannot by itself deliver visible accountability for outbound action. A decision on this case needs the team to draw a workflow showing draft, evidence validation, human approval and outbound-send controls and verify that outcome.
C:This shifts attention toward communicating justified regional deviations when the relevant decision is about visible accountability for outbound action. Unless the architecture can draw a workflow showing draft, evidence validation, human approval and outbound-send controls, the failure described in the stem remains unresolved.
Question 331
Two departments share a Claude service and dispute how inference and connector expenses should be allocated. The platform reports only total spend, and neither department can see how its requests drive costs. What should the architect present to budget owners?
- Present a transparent cost-allocation decision using attributable per-department usage, shared-overhead rules and a documented chargeback model.
- Divide every cost equally between departments, irrespective of traffic, tool use or shared infrastructure.
- Assign all costs to whichever department launched the original proof of concept.
- Show only the price per input token because connector and review costs are outside the model billing line.
Explanation
Correct Answer
A:A combined invoice hides which workloads generate variable spend. Request and connector attribution, agreed allocation treatment for common services, and a reproducible example let sponsors evaluate fair funding alternatives.
Incorrect Answers
B:An arbitrary 50–50 split can subsidize disproportionate usage and discourages accountability. Budget owners need to agree how measurable variable costs and genuinely shared overhead are handled.
C:Historical sponsorship does not establish who consumes current production capacity. An attribution and allocation policy should follow actual use and agreed organizational funding rules.
D:Per-token price cannot allocate the total service bill or reflect department-specific connector consumption. The decision requires an attributable service-level cost model.
Question 332
An architecture committee is deciding between one shared search gateway and separate departmental information APIs. Security favors isolation, while operations favors centralized maintenance. What artifact best communicates the tradeoff and supports an accountable decision?
- Assert that a centralized gateway is inherently secure and does not require per-user authorization tests.
- Select departmental APIs solely because they were deployed first, without comparing future operating obligations.
- Present only the number of services in each option and treat fewer services as sufficient evidence.
- Present an option-by-option decision record comparing entitlement boundaries, operational ownership, migration effort and failure blast radius.
Explanation
Correct Answer
D:Shared and separate integration designs place enforcement and operational burdens in different places. A decision record with explicit criteria, assumptions and risk owners makes the tradeoffs reviewable without pretending that one design always wins.
Incorrect Answers
A:Centralization does not remove document-level entitlements or the impact of a compromised gateway. The committee still needs verified boundaries and a reasoned comparison.
B:Deployment age is not an architecture acceptance criterion. Ownership, isolation requirements and integration complexity must be evaluated against current business needs.
C:A component count cannot reveal blast radius, entitlement enforcement or support accountability. The decision needs evaluated risks and operational responsibilities.
Question 333
A security committee objects to enterprise search integration because it assumes all users can see all indexed files. Which intervention most directly addresses the failure?
- Describe region-specific decision constraints and a common control baseline.
- Demonstrate authorization filters and identity propagation in the actual data flow.
- Separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
- Explain coordination risks and propose an evaluated simple baseline before adding agents.
Explanation
Correct Answer
B:Concrete access-control boundaries clarify where retrieval enforces existing permissions. Show evidence for verifiable document entitlement design alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
A:For a workflow whose primary weakness was communicating justified regional deviations, that proposal could be reasonable. Here the unmet obligation is verifiable document entitlement design, and that requires the system to demonstrate authorization filters and identity propagation in the actual data flow.
C:It is aimed at a different outcome, transparent evidence and assumptions. No evidence is given that it provides verifiable document entitlement design; the needed mechanism is to demonstrate authorization filters and identity propagation in the actual data flow at the appropriate boundary.
D:It emphasizes evidence-led multi-agent architecture choice without closing the stated gap in verifiable document entitlement design. By contrast, the remedy to demonstrate authorization filters and identity propagation in the actual data flow acts on the condition that makes this scenario fail.
Question 334
A delivery team wants to change models immediately after a vendor release. What should the solution architect prioritize?
- Document upgrade benefits, validation scope, operational risks and rollback triggers.
- Describe region-specific decision constraints and a common control baseline.
- Present a decision record with measured quality, latency, total cost and residual risk.
- Show cost per accepted case including inference, rework, human review and failures.
Explanation
Correct Answer
A:A change decision record protects release quality while acknowledging innovation opportunities. Show evidence for accountable model change management alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
B:The alternative could reduce risk associated with communicating justified regional deviations, yet the described behavior requires accountable model change management. That outcome depends on the capability to document upgrade benefits, validation scope, operational risks and rollback triggers.
C:The proposed measure is aimed at transparent model-tier tradeoff decision, not accountable model change management. The architecture must instead document upgrade benefits, validation scope, operational risks and rollback triggers; a separate change is still needed for accountable model change management.
D:The relevant trade-off is not primarily whole-process cost comparison. A successful fix needs accountable model change management; otherwise accountable model change management remains unverified even after this change.
Question 335
A stakeholder asks whether a prompt is enough to stop a bot from issuing refunds. Which design choice best matches the stated constraint?
- Draw a workflow showing draft, evidence validation, human approval and outbound-send controls.
- Describe abstention, escalation and source gaps using examples of acceptable failure behavior.
- Show the distinction between behavioral instructions and gateway-enforced transaction permissions.
- Document upgrade benefits, validation scope, operational risks and rollback triggers.
Explanation
Correct Answer
C:Communicating defense in depth prevents false assurance from model-level policy text. Show evidence for clear explanation of enforceable safeguards alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
A:Its main effect concerns visible accountability for outbound action. The case explicitly hinges on clear explanation of enforceable safeguards; to resolve it reliably, the implementation has to show the distinction between behavioral instructions and gateway-enforced transaction permissions, which this choice does not establish.
B:The option might improve communicated safe failure modes; that remains insufficient for clear explanation of enforceable safeguards. The case is resolved by clear explanation of enforceable safeguards, which depends on a design that will show the distinction between behavioral instructions and gateway-enforced transaction permissions.
D:The design concern is real but different: accountable model change management. It does not impose the specific requirement for clear explanation of enforceable safeguards; a suitable solution must show the distinction between behavioral instructions and gateway-enforced transaction permissions and demonstrate the result.
Question 336
A technical steering group disagrees over whether to use RAG or fine-tuning for fast-changing internal policies. Which architecture adjustment is most appropriate?
- Present a decision record with measured quality, latency, total cost and residual risk.
- Separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
- Compare update frequency, evidence traceability and data ownership in a decision matrix.
- Demonstrate authorization filters and identity propagation in the actual data flow.
- Describe abstention, escalation and source gaps using examples of acceptable failure behavior.
Explanation
Correct Answer
C:Retrieval often fits changing source knowledge; adaptation may serve other needs such as consistent format or behavior.
Incorrect Answers
A:It emphasizes transparent model-tier tradeoff decision without closing the stated gap in fit-based retrieval-versus-customization reasoning. By contrast, the remedy to compare update frequency, evidence traceability and data ownership in a decision matrix acts on the condition that makes this scenario fail.
B:That is primarily a technique for transparent evidence and assumptions. It lacks the explicit measure for fit-based retrieval-versus-customization reasoning that is necessary here, namely a system that will compare update frequency, evidence traceability and data ownership in a decision matrix.
D:This recommendation targets verifiable document entitlement design at a different point in the workflow. It leaves fit-based retrieval-versus-customization reasoning outside the design contract and does not require the team to compare update frequency, evidence traceability and data ownership in a decision matrix.
E:This is a valid architectural concern when optimizing communicated safe failure modes. Here it leaves fit-based retrieval-versus-customization reasoning unaddressed, whereas the intended change must compare update frequency, evidence traceability and data ownership in a decision matrix before the case can be considered resolved.
Question 337
A business owner wants a guaranteed answer for every unknown policy question. Which approach should be recommended?
- Describe abstention, escalation and source gaps using examples of acceptable failure behavior.
- Separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
- Compare update frequency, evidence traceability and data ownership in a decision matrix.
- Show cost per accepted case including inference, rework, human review and failures.
Explanation
Correct Answer
A:Uncertainty handling is a product requirement rather than evidence of poor engineering. Show evidence for communicated safe failure modes alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
B:The control would be justified if transparent evidence and assumptions were the limiting factor. In this scenario the decisive target is communicated safe failure modes; success requires an architecture that will describe abstention, escalation and source gaps using examples of acceptable failure behavior.
C:The option belongs to the broader work on fit-based retrieval-versus-customization reasoning, but it cannot by itself deliver communicated safe failure modes. A decision on this case needs the team to describe abstention, escalation and source gaps using examples of acceptable failure behavior and verify that outcome.
D:This shifts attention toward whole-process cost comparison when the relevant decision is about communicated safe failure modes. Unless the architecture can describe abstention, escalation and source gaps using examples of acceptable failure behavior, the failure described in the stem remains unresolved.
Question 338
A supplier's demo lacks the production load and tenant-isolation data required by procurement. What is the best production design response?
- Present a decision record with measured quality, latency, total cost and residual risk.
- Separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
- Explain coordination risks and propose an evaluated simple baseline before adding agents.
- Describe abstention, escalation and source gaps using examples of acceptable failure behavior.
Explanation
Correct Answer
B:Credible architecture communication identifies what is evidenced and what remains conditional. Show evidence for transparent evidence and assumptions alongside alternatives and residual assumptions so the decision can be challenged before commitment.
Incorrect Answers
A:The design concern is real but different: transparent model-tier tradeoff decision. It does not impose the specific requirement for transparent evidence and assumptions; a suitable solution must separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register and demonstrate the result.
C:That choice would be responsive to evidence-led multi-agent architecture choice, but the reported defect is different. It leaves transparent evidence and assumptions unproved because there is no commitment to separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
D:There is a legitimate use case for communicated safe failure modes, but it does not supply the test or enforcement needed for transparent evidence and assumptions. The design needs to separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
Question 339
An international organization needs documented reasons for deviating from a common architecture where local data laws differ. Its steering committee also sees only global averages that hide poor answer quality in a small region. Which TWO communication deliverables address the objections? Choose TWO.
- Present a decision record with measured quality, latency, total cost and residual risk.
- Explain coordination risks and propose an evaluated simple baseline before adding agents.
- Present region-stratified quality and risk evidence with local acceptance thresholds and an owner for each exception.
- Separate demonstrated capabilities from assumptions, gaps and planned tests in a risk register.
- Describe region-specific decision constraints and a common control baseline.
Explanation
Correct Answers
C:An overall mean cannot demonstrate service suitability in every jurisdiction. Showing the tested distribution, agreed thresholds and accountable exceptions makes the steering decision inspectable rather than a global reassurance.
E:Regional exceptions require explicit reasons tied to local rules and a shared baseline of controls. A decision record should show where the architecture diverges and which authority accepted the exception; regional performance evidence provides the complementary operating justification.
Incorrect Answers
A:This recommendation targets transparent model-tier tradeoff decision at a different point in the workflow. It leaves communicating justified regional deviations and decision-ready regional performance evidence outside the design contract and does not require the team to describe region-specific decision constraints and a common control baseline.
B:This is a valid architectural concern when optimizing evidence-led multi-agent architecture choice. Here it leaves communicating justified regional deviations and decision-ready regional performance evidence unaddressed, whereas the intended change must present region-stratified quality and risk evidence with local acceptance thresholds and an owner for each exception before the case can be considered resolved.
D:By itself, this improves transparent evidence and assumptions, but the case is not asking the architect to optimize that dimension. The requirement is communicating justified regional deviations and decision-ready regional performance evidence, supported by an approach to present region-stratified quality and risk evidence with local acceptance thresholds and an owner for each exception.
Question 340
A sales presentation promises that a support assistant can keep and search unlimited uploaded documents within a single chat. Actual context and ingestion limits require durable indexing of large files. How should the architect reset the service expectation?
- Promise unlimited context anyway because the application can retry model calls as often as needed.
- Disable upload-size alerts but continue advertising that every attachment remains in model memory forever.
- Describe model output quality only and exclude indexing or document lifecycle behavior from the service agreement.
- Distinguish conversation context from durable document retrieval and publish tested file limits, indexing behavior and fallback commitments.
Explanation
Correct Answer
D:A chat context window is not an unlimited document repository. Communicating tested ingestion sizes, retrieval coverage and what happens before an index is ready creates a contract that the application can actually deliver.
Incorrect Answers
A:Retrying does not remove context-window or ingestion constraints. The product needs an explicit supported capacity and a defined large-file workflow.
B:Suppressing warnings makes an unsupported promise harder to detect. Durable storage and retrieval must be separately designed and disclosed.
C:Answer fluency cannot guarantee that an unindexed or expired document is retrievable. Input capacity and readiness are observable parts of the user-facing service commitment.