Question 210
Implementation alternatives must be compared for a workflow where the model submits malformed identifiers to an operational API. The pilot shows this gap: Invalid record IDs produce updates to the wrong business object. Which alternative should lead the solution?
- Build a bespoke agent when differentiated orchestration is essential
- Use staged rollout with feature flags and pilot cohorts
- Validate generated parameters against typed business schemas
- Require approval before irreversible agent actions
Explanation
Correct Answer
C:Because typed inputs and server validation reject malformed or out-of-scope calls, this approach corrects the reported breakdown where invalid record IDs produce updates to the wrong business object; the affected situation is that the model submits malformed identifiers to an operational API.
Incorrect Answers
A:There is a valid place for the mechanism that custom engineering trades more control for more operating responsibility, namely competitive requirements depend on unique multi-model control; it does not remove the specific defect that invalid record IDs produce updates to the wrong business object from this case.
B:For new behavior must be evaluated without exposing the full workforce, this other approach relies on the fact that controlled release gates restrict blast radius and support rollback; the issue here is invalid record IDs produce updates to the wrong business object; that mechanism does not establish the required typed inputs and server validation reject malformed or out-of-scope calls.
D:For another type of architecture, confirmation and human approval separate advice from execution authority is useful; this design must instead support the model submits malformed identifiers to an operational API despite the problem where invalid record IDs produce updates to the wrong business object.
Question 211
Risk reviewers need a control, not merely a monitoring dashboard. The use case is: Prompt revisions cause intermittent regressions. Potential harm: A maker might publish a candidate despite a failed critical evaluation gate. Which protection fits?
- Require explicit approval when critical thresholds fail
- Alert on meaningful user failures, not just raw token volume
- Enforce owner sign-off for high-risk grounding collections
- Treat generated expectations as suggestions until verified
Explanation
Correct Answer
A:In the scenario where prompt revisions cause intermittent regressions, the explicit risk is that a maker might publish a candidate despite a failed critical evaluation gate; enforcement must follow the mechanism by which stable automated suites compare response quality across versions.
Incorrect Answers
B:This restriction is associated with operations must distinguish healthy traffic from quality incidents and addresses the exposure that alerting might fire on raw request volume while missing user-impacting failures; it does not specifically protect the target workflow against a maker might publish a candidate despite a failed critical evaluation gate.
C:This restriction is associated with new documents enter a production source automatically and addresses the exposure that sensitive ownerless documents might enter production grounding; it does not specifically protect the target workflow against a maker might publish a candidate despite a failed critical evaluation gate.
D:The safeguard for agents can call APIs with customer IDs and amount values deals with a different threat: a hallucinated expected outcome might become a trusted automated oracle; even after it is enforced, the exposure that a maker might publish a candidate despite a failed critical evaluation gate remains possible in the target workflow where stable automated suites compare response quality across versions.
Question 212
Risk reviewers need a control, not merely a monitoring dashboard. The use case is: Shared result caching makes answers fast across roles. Potential harm: A user might receive another principal's cached protected answer. Which protection fits?
- Prevent sensitive data from entering prompts unless required
- Revoke shared credentials and forbid cross-stage access
- Block publishing when required evidence cannot be recovered
- Do not share private cached outputs across principals
Explanation
Correct Answer
D:Shared result caching makes answers fast across roles; its risk scenario is that a user might receive another principal's cached protected answer; the safeguard fits because cache keys and access validation prevent security-context leakage and the restriction addresses that specific exposure.
Incorrect Answers
A:This restriction is associated with an internal model sometimes returns confidential text verbatim and addresses the exposure that a model might expose sensitive user data in uncontrolled prompt output; it does not specifically protect the target workflow against a user might receive another principal's cached protected answer.
B:The alternate guard addresses staging credentials might allow writes to production business systems for test automation must not reach live customer records; the present threat, a user might receive another principal's cached protected answer, needs a different enforcement boundary.
C:For the use case where an answer contradicts the handbook despite high model fluency, the control addresses the threat that the team might publish a model that never retrieves the evidence it requires; the specified agent must be protected against a user might receive another principal's cached protected answer.
Question 213
The architecture board has evidence that scanned forms suddenly produce inaccurate predictions. The affected situation is: Production documents differ from clean training samples. What change most directly establishes the missing behavior?
- Retain rollback snapshots for models and data transformations
- Use Prompt Shields or equivalent intervention-point screening
- Assess robustness under realistic drift and noisy input
- Validate custom extraction models by field and document class
Explanation
Correct Answer
C:For the need that production documents differ from clean training samples, the relevant mechanism is that stress tests measure sensitivity to expected input variation; that architectural dependency addresses the observed failure where scanned forms suddenly produce inaccurate predictions.
Incorrect Answers
A:In the case where a preprocessing release changes category labels incorrectly, the alternate design depends on versioned data transformations and model artifacts enable consistent rollback; it does not correct the target failure, scanned forms suddenly produce inaccurate predictions.
B:Input and document-attack detection supplies layered protection; its main purpose is hostile instructions arrive both as user input and retrieved documents, not correction of the reported defect where scanned forms suddenly produce inaccurate predictions in the target process.
D:This design would make sense if a document model performs well overall but fails on tax identifiers, particularly where a high aggregate score masks critical field errors; the requested architecture instead needs the dependency that stress tests measure sensitivity to expected input variation because scanned forms suddenly produce inaccurate predictions.
Question 214
A solution architecture is being reviewed for this business situation: Many calls require only a small fixed-label output. The observed defect is that universal use of a premium model inflates cost. Which architectural decision addresses the underlying requirement?
- Route simple classification requests to a validated inexpensive model
- Create a custom-engine agent for bespoke model orchestration
- Route high-stakes reasoning to a stronger validated model
- Include human review and exception handling in unit economics
Explanation
Correct Answer
A:For the need that many calls require only a small fixed-label output, the relevant mechanism is that task-aware routing chooses economical models after quality gates; that architectural dependency addresses the observed failure where universal use of a premium model inflates cost.
Incorrect Answers
B:There is a valid place for the mechanism that custom-engine architecture controls runtime, model calls and integrations, namely the solution demands its own model selection and hosting policies; it does not remove the specific defect that universal use of a premium model inflates cost from this case.
C:For some cases need cross-document reasoning and detailed justification, this other approach relies on the fact that complexity classification selects a capable model for difficult tasks; the issue here is universal use of a premium model inflates cost; that mechanism does not establish the required task-aware routing chooses economical models after quality gates.
D:For AI drafts requests quickly but specialists must correct some outputs, this other approach relies on the fact that end-to-end unit cost includes review, escalation and retry effort; the issue here is universal use of a premium model inflates cost; that mechanism does not establish the required task-aware routing chooses economical models after quality gates.
Question 215
The operating team must monitor whether an architectural correction works. Business context: A support agent changes how it classifies priority intents. Historical symptom: High-severity tickets begin routing to general queues. Which outcome measure is the best match?
- Production calls routed to approved production dependencies
- Customer handoffs free from integration mismatches
- Environment promotions without configuration drift
- Critical cases assigned to the correct escalation group
Explanation
Correct Answer
D:The use case involves a support agent changes how it classifies priority intents and the baseline failure that high-severity tickets begin routing to general queues; the KPI tracks critical cases assigned to the correct escalation group, so improvement has an interpretable operational meaning.
Incorrect Answers
A:For the situation where a flow moved to production still points to a test endpoint, production calls routed to approved production dependencies is a sensible indicator; the problem that high-severity tickets begin routing to general queues requires measurement of critical cases assigned to the correct escalation group.
B:For the situation where a customer handoff depends on shared record and term mappings, customer handoffs free from integration mismatches is a sensible indicator; the problem that high-severity tickets begin routing to general queues requires measurement of critical cases assigned to the correct escalation group.
C:For the workflow where the same agent must move from development to test and production, environment promotions without configuration drift is an actionable measure; it does not establish whether high-severity tickets begin routing to general queues in the situation under review.
Question 216
The acceptance board asks how to verify a corrected design. The setting is: A conversational change affects only one support geography. Last time, global averages hide a regional escalation spike. Which scenario should the team execute?
- Attempt an unapproved production modification in test governance
- Exercise the target channel after a staged promotion
- Promote to staging and confirm correct connection targets
- Invoke an agent after deployment and locate its trace
- Roll out to one region and check segmented resolution metrics
Explanation
Correct Answer
E:For the required workflow where a conversational change affects only one support geography, the test needs to reproduce or prevent the failure where global averages hide a regional escalation spike; rolling out to one region and checking segmented resolution metrics exercises the path where cohort telemetry supports safe staged validation.
Incorrect Answers
A:The verification concerning production components should not be edited casually checks managed solution deployment supports controlled downstream customization boundaries against a maker changes production agent logic outside the release process; its result says little about the observed issue where global averages hide a regional escalation spike.
B:The tested behavior is that post-import validation includes publishing and channel behavior; the necessary evidence here concerns cohort telemetry supports safe staged validation and the previously observed global averages hide a regional escalation spike.
C:Promoting to staging and confirming correct connection targets verifies an adjacent objective where a flow moved to production still points to a test endpoint; it leaves untested the causal path in which cohort telemetry supports safe staged validation for this case.
D:Although the exercise might catch the on-call team cannot explain failed tool calls, it would not directly validate the mechanism that cohort telemetry supports safe staged validation in the target implementation.
Question 217
An enterprise team is making a platform design decision. Business context: Personal read-only assistants and payment agents have different risk. Present failure: Every agent receives the same minimal review. Which approach resolves the mismatch?
- Use progressive deployment and rollback for a stable agent endpoint
- Restrict who can supply tuning data and approve model updates
- Apply risk-tiered approval gates to agent publication
- Define policies for third-party agent onboarding
Explanation
Correct Answer
C:The mechanism that zoned governance aligns controls with impact and data sensitivity fits the scenario where personal read-only assistants and payment agents have different risk; it addresses the defect that every agent receives the same minimal review.
Incorrect Answers
A:The alternative addresses a model upgrade breaks downstream response formatting by a design in which canary traffic and versioned rollback reduce release risk; it does not supply the missing relationship for personal read-only assistants and payment agents have different risk, where every agent receives the same minimal review.
B:For another type of architecture, separated data-curation and release permissions preserve integrity is useful; this design must instead support personal read-only assistants and payment agents have different risk despite the problem where every agent receives the same minimal review.
D:The alternative addresses an unreviewed plugin exports sensitive tenant information by a design in which procurement and security review evaluate permissions and data flows; it does not supply the missing relationship for personal read-only assistants and payment agents have different risk, where every agent receives the same minimal review.
Question 218
A quality engineer is writing a regression case after the incident where an upstream field rename silently breaks downstream grounding. The underlying business scenario is: Multiple AI teams consume a changing source API. What should the test exercise?
- Query immediately before and after an effective-date change
- Emit duplicate and delayed events and verify idempotent processing
- Run contract tests against both current and previous schema versions
- Re-run a saved golden set across prompt versions
Explanation
Correct Answer
C:The scenario is that multiple AI teams consume a changing source API; the defect to rule out is that an upstream field rename silently breaks downstream grounding; testing via running contract tests against both current and previous schema versions challenges the design dependency where versioned schemas and compatibility policy make interfaces predictable.
Incorrect Answers
A:In a solution where policy documents contain draft, superseded, and currently effective revisions, querying immediately before and after an effective-date change could catch the defect that the agent cites a superseded rule despite a newer policy existing; the current release gate must challenge the failure where an upstream field rename silently breaks downstream grounding.
B:The suggested test serves invoice holds must trigger exception handling without polling, in which delayed reconciliation hides material fulfillment blockers; the present scenario instead concerns the reported problem where an upstream field rename silently breaks downstream grounding.
D:The verification concerning many teams copy slightly different instructions for the same task checks governed templates provide versioned contracts and test fixtures against a successful prompt change cannot be reproduced after rollout; its result says little about the observed issue where an upstream field rename silently breaks downstream grounding.
Question 219
The fault affects an AI-powered workflow in which a candidate appears accurate on its training examples. Specifically, overfitting hides failures on new business records. Which diagnostic investigation best distinguishes the relevant failure path?
- Inspect transcript alignment and transcription errors
- Inspect target UI changes and automation failure traces
- Compare training scores with independently curated validation results
- Review agent capability manifest against requested functions
Explanation
Correct Answer
C:For the case where a candidate appears accurate on its training examples, the diagnostic path must explain the failure that overfitting hides failures on new business records; comparing training scores with independently curated validation results targets the mechanism by which separation of training and evaluation yields realistic generalization checks.
Incorrect Answers
A:Inspecting transcript alignment and transcription errors explores a separate connection, where speech transcription transforms audio to text for subsequent analysis; the diagnosed requirement is a candidate appears accurate on its training examples, and the actual defect is overfitting hides failures on new business records.
B:Inspecting target ui changes and automation failure traces explores a separate connection, where uI automation can operate application interfaces but requires stability checks; the diagnosed requirement is a candidate appears accurate on its training examples, and the actual defect is overfitting hides failures on new business records.
D:The proposed inspection is associated with employees want curated assistance without owning a separate LLM runtime, where declarative agents define instructions, knowledge and actions in the supported host; it would not expose the failure mode that overfitting hides failures on new business records with the required specificity.
Question 220
Before production promotion, the team needs evidence for this setting: The system cannot always find evidence for regulated guidance. A prior regression caused the following: An agent guesses a policy instead of escalating. Which validation is most persuasive?
- Submit edge cases and validate response shape
- Remove required sources and check that the agent declines
- Run multi-hop evaluation cases for the strong-model route
- Attempt low-risk and high-risk actions and inspect approval enforcement
Explanation
Correct Answer
B:The scenario is that the system cannot always find evidence for regulated guidance; the defect to rule out is that an agent guesses a policy instead of escalating; testing via removing required sources and checking that the agent declines challenges the design dependency where confidence and evidence checks route insufficiently supported cases to humans.
Incorrect Answers
A:For agents return inconsistent fields for downstream integrations, the alternate test is about whether clear instructions and structured outputs reduce avoidable ambiguity; it cannot prove the outcome needed after the reported regression where an agent guesses a policy instead of escalating.
C:The tested behavior is that complexity classification selects a capable model for difficult tasks; the necessary evidence here concerns confidence and evidence checks route insufficiently supported cases to humans and the previously observed an agent guesses a policy instead of escalating.
D:A pass on that verification would speak to teams propose that an agent perform both low-risk lookups and financial postings; it would not show that confidence and evidence checks route insufficiently supported cases to humans, which is needed to avoid the observed defect an agent guesses a policy instead of escalating.