Question 140
After implementing a workflow to apply input and output harm filters, the team needs a robust release gate. Which check should it choose?
- Sample logs to verify redaction coverage and authorized retention behavior
- Induce a circular task and verify deterministic termination and budget reporting
- Run a policy-labeled evaluation set measuring block rates and legitimate-request false positives
- Red-team malicious instructions inside retrieved pages and record detected unsafe tool intentions
Explanation
Correct Answer
C:The accepted change is to apply input and output harm filters; this evaluation checks whether it can limit harmful user prompts and unsafe assistant responses according to an approved safety policy, since content filters classify supported harm categories at the inference boundary; thresholds require evaluation for false positives and misses.
Incorrect Answers
A:This test concerns the ability to keep traces useful without retaining customer identifiers or secrets in operational logs, for which PII-aware redaction and minimal logging reduce privacy risk; collecting full prompts in diagnostic systems is not inherently necessary; it does not establish whether the change to apply input and output harm filters works for the current release.
B:This test concerns the ability to prevent infinite tool loops and uncontrolled consumption in long-running workflows, for which execution limits on tool calls, runtime and cost bound what autonomy can do, independent of prompt compliance; it does not establish whether the change to apply input and output harm filters works for the current release.
D:This test concerns the ability to prevent an untrusted document from instructing an agent to ignore its governing rules, for which prompt Shields and related controls can detect indirect instructions embedded in lower-trust content; document text remains evidence, not authority; it does not establish whether the change to apply input and output harm filters works for the current release.
Question 141
An operator can reproduce this error: A container upgrade points its inference client to a nonproduction deployment left in source code. Which targeted repair is justified?
- Retrieve current authorized passages and cite them when forming the response
- Load approved project and deployment settings from the production environment
- Configure a credential chain using the workload identity and least-privilege role grants
- Reject invalid parameters with application-side validation before the business API call
Explanation
Correct Answer
B:The fault signal is that a container upgrade points its inference client to a nonproduction deployment left in source code; the targeted fix works because endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project, restoring the ability to avoid routing production questions to a test model through hardcoded SDK settings.
Incorrect Answers
A:If the failing function were to answer questions about changing internal procedures with source-backed responses, this change could be relevant because RAG retrieves relevant evidence at runtime and gives it to a generative model; model pretraining alone cannot guarantee current policy text; the actual symptom is that a container upgrade points its inference client to a nonproduction deployment left in source code, which calls for a different repair.
C:If the failing function were to authenticate a server-side generative app through Entra where supported, this change could be relevant because a workload identity and appropriate RBAC reduce static credential exposure while preserving auditable authorization decisions; the actual symptom is that a container upgrade points its inference client to a nonproduction deployment left in source code, which calls for a different repair.
D:If the failing function were to prevent malformed model-generated parameters from reaching a write-capable API, this change could be relevant because tool schemas guide argument shape, but the application must check types, constraints, identity and authorization before execution; the actual symptom is that a container upgrade points its inference client to a nonproduction deployment left in source code, which calls for a different repair.
Question 142
A new Foundry-based application must catch plausible statements that are not substantiated by the retrieved evidence. Which approach should the team select first?
- Use comparative evaluations for prompt changes
- Add bounded tool-use loops for multi-step tasks
- Use passage-level citations in RAG answers
- Measure factual faithfulness to source context
- Use a multimodal request for document-plus-image reasoning
- Evaluate answer relevance against user intent
Explanation
Correct Answer
D:For the stated requirement to catch plausible statements that are not substantiated by the retrieved evidence, faithfulness checks whether generated claims follow provided source material, not whether wording is eloquent; this is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is to choose between candidate prompt versions by measured task quality rather than intuition, paired evaluation on identical cases isolates prompt impact; production traffic shifts make uncontrolled comparisons misleading; this does not directly meet the required outcome to catch plausible statements that are not substantiated by the retrieved evidence.
B:For a system whose objective is to allow a model to request multiple independent data lookups while limiting cost and duration, an orchestration loop can inspect tool requests, execute authorized tools, return results, and terminate under explicit limits; this does not directly meet the required outcome to catch plausible statements that are not substantiated by the retrieved evidence.
C:For a system whose objective is to let a compliance reviewer locate the specific evidence supporting each generated assertion, citations tied to retrieved chunks improve traceability; a source list without claim alignment does not prove each claim; this does not directly meet the required outcome to catch plausible statements that are not substantiated by the retrieved evidence.
E:For a system whose objective is to reason about a photo and related user instructions together in one supported inference flow, a multimodal model can interpret visual input alongside textual context; relying on OCR alone loses non-textual relationships; this does not directly meet the required outcome to catch plausible statements that are not substantiated by the retrieved evidence.
F:For a system whose objective is to detect answers that are grounded in retrieved text but do not address the actual question, relevance evaluates how directly a response serves the request; factual grounding and response usefulness are related but distinct axes; this does not directly meet the required outcome to catch plausible statements that are not substantiated by the retrieved evidence.
Question 143
An Azure AI workload passes basic health probes but has this issue: Answers fail when a condition begins at the end of one chunk and its exception starts in the next. Which fix targets it?
- Detect empty or low-confidence retrieval and return a supported no-answer or escalation response
- Adjust chunk boundaries and limited overlap, then rebuild affected embeddings
- Provide image content to a vision-capable model with a bounded question and verification rubric
- Filter authorized indexed chunks using tenant, department, and effective-date fields
- Execute independent authorized reads in parallel and join results before generation
Explanation
Correct Answer
B:The fault signal is that answers fail when a condition begins at the end of one chunk and its exception starts in the next; the targeted fix works because chunk overlap can retain nearby context across splits, although excessive overlap inflates index size and duplicates retrieval, restoring the ability to keep context that crosses page or section boundaries from being lost during indexing.
Incorrect Answers
A:If the failing function were to avoid fabricating an answer to a question not covered by the organization's knowledge collection, this change could be relevant because an evidence threshold and abstention path distinguish unknown facts from model completion skill; confident text is not proof of source support; the actual symptom is that answers fail when a condition begins at the end of one chunk and its exception starts in the next, which calls for a different repair.
C:If the failing function were to reason about a photo and related user instructions together in one supported inference flow, this change could be relevant because a multimodal model can interpret visual input alongside textual context; relying on OCR alone loses non-textual relationships; the actual symptom is that answers fail when a condition begins at the end of one chunk and its exception starts in the next, which calls for a different repair.
D:If the failing function were to ensure answers include only documents for a user's department and effective policy year, this change could be relevant because retrieval filters narrow the evidence set by structured attributes before the model sees it; prompt instructions are not security filters; the actual symptom is that answers fail when a condition begins at the end of one chunk and its exception starts in the next, which calls for a different repair.
E:If the failing function were to reduce latency when separate read-only lookups have no ordering dependency, this change could be relevant because parallelizing independent operations shortens critical-path latency but is unsafe when one action requires another's output; the actual symptom is that answers fail when a condition begins at the end of one chunk and its exception starts in the next, which calls for a different repair.
Question 144
An operator can reproduce this error: An invoice draft is generated before customer eligibility has been checked. Which targeted repair is justified?
- Execute independent authorized reads in parallel and join results before generation
- Validate eligibility first, run generation second, and require approval before sending
- Use deterministic status rules for routing and reserve the LLM for language-specific subtasks
- Load approved project and deployment settings from the production environment
Explanation
Correct Answer
B:The fault signal is that an invoice draft is generated before customer eligibility has been checked; the targeted fix works because explicit workflow stages make ordering and failure handling reproducible; free-form agent planning is unnecessary for a fixed process, restoring the ability to execute a known approval workflow where inputs must be validated before a model drafts a response.
Incorrect Answers
A:If the failing function were to reduce latency when separate read-only lookups have no ordering dependency, this change could be relevant because parallelizing independent operations shortens critical-path latency but is unsafe when one action requires another's output; the actual symptom is that an invoice draft is generated before customer eligibility has been checked, which calls for a different repair.
C:If the failing function were to choose among distinct business flows based on a validated account status, this change could be relevant because rules engines offer predictable decisions for simple structured conditions, while language models handle unstructured interpretation; the actual symptom is that an invoice draft is generated before customer eligibility has been checked, which calls for a different repair.
D:If the failing function were to avoid routing production questions to a test model through hardcoded SDK settings, this change could be relevant because endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project; the actual symptom is that an invoice draft is generated before customer eligibility has been checked, which calls for a different repair.
Question 145
The team plans to handle streamed generation with cancellation. Which test provides the clearest evidence that this change works?
- Inspect request-target metadata and run an environment-isolation smoke test
- Test for cross-department leakage and expired-document citation rate
- Prove the app can generate responses after removing the static key from configuration
- Measure cancellation propagation and discarded tokens after client disconnects
Explanation
Correct Answer
D:The accepted change is to handle streamed generation with cancellation; this evaluation checks whether it can show partial assistant output to a user while stopping work after the request is canceled, since streaming delivers incremental output; cancellation and backpressure prevent orphaned inference work and stale interface content.
Incorrect Answers
A:This test concerns the ability to avoid routing production questions to a test model through hardcoded SDK settings, for which endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project; it does not establish whether the change to handle streamed generation with cancellation works for the current release.
B:This test concerns the ability to ensure answers include only documents for a user's department and effective policy year, for which retrieval filters narrow the evidence set by structured attributes before the model sees it; prompt instructions are not security filters; it does not establish whether the change to handle streamed generation with cancellation works for the current release.
C:This test concerns the ability to authenticate a server-side generative app through Entra where supported, for which a workload identity and appropriate RBAC reduce static credential exposure while preserving auditable authorization decisions; it does not establish whether the change to handle streamed generation with cancellation works for the current release.
Question 146
During a regression test, the following unexpected result occurs: A valid tool contract works in demos but a missing optional field causes an uncaught exception. What should the engineer do?
- Evaluate claim support against the supplied passages and require repair or abstention on failures
- Add malformed-input fixtures and explicit parser error handling to the integration tests
- Filter authorized indexed chunks using tenant, department, and effective-date fields
- Add question-answer relevance evaluations with labeled examples and failure review
Explanation
Correct Answer
B:The fault signal is that a valid tool contract works in demos but a missing optional field causes an uncaught exception; the targeted fix works because boundary-case testing surfaces parsing and orchestration failures that standard benchmark prompts miss, restoring the ability to verify that an app recovers safely from empty messages, long context, and invalid tool responses.
Incorrect Answers
A:If the failing function were to catch plausible statements that are not substantiated by the retrieved evidence, this change could be relevant because faithfulness checks whether generated claims follow provided source material, not whether wording is eloquent; the actual symptom is that a valid tool contract works in demos but a missing optional field causes an uncaught exception, which calls for a different repair.
C:If the failing function were to ensure answers include only documents for a user's department and effective policy year, this change could be relevant because retrieval filters narrow the evidence set by structured attributes before the model sees it; prompt instructions are not security filters; the actual symptom is that a valid tool contract works in demos but a missing optional field causes an uncaught exception, which calls for a different repair.
D:If the failing function were to detect answers that are grounded in retrieved text but do not address the actual question, this change could be relevant because relevance evaluates how directly a response serves the request; factual grounding and response usefulness are related but distinct axes; the actual symptom is that a valid tool contract works in demos but a missing optional field causes an uncaught exception, which calls for a different repair.
Question 147
After implementing a workflow to use function calling for external calculations, the team needs a robust release gate. Which check should it choose?
- Score relationship questions on manually labeled image-and-text examples
- Compute unsupported-claim rate on source-grounded summaries
- Compare calculated totals with independently tested business-rule outputs
- Measure answer support and document freshness on questions with recently revised policies
Explanation
Correct Answer
C:The accepted change is to use function calling for external calculations; this evaluation checks whether it can obtain exact tax or pricing totals from an authoritative business calculation endpoint, since a tool call delegates deterministic computation to trusted code instead of relying on a model's arithmetic approximation.
Incorrect Answers
A:This test concerns the ability to reason about a photo and related user instructions together in one supported inference flow, for which a multimodal model can interpret visual input alongside textual context; relying on OCR alone loses non-textual relationships; it does not establish whether the change to use function calling for external calculations works for the current release.
B:This test concerns the ability to catch plausible statements that are not substantiated by the retrieved evidence, for which faithfulness checks whether generated claims follow provided source material, not whether wording is eloquent; it does not establish whether the change to use function calling for external calculations works for the current release.
D:This test concerns the ability to answer questions about changing internal procedures with source-backed responses, for which RAG retrieves relevant evidence at runtime and gives it to a generative model; model pretraining alone cannot guarantee current policy text; it does not establish whether the change to use function calling for external calculations works for the current release.
Question 148
A development group needs its AI workflow to execute a known approval workflow where inputs must be validated before a model drafts a response. What technical approach most directly enables that?
- Run independent tool requests concurrently where safe
- Add bounded tool-use loops for multi-step tasks
- Orchestrate a deterministic sequence around model steps
- Use comparative evaluations for prompt changes
Explanation
Correct Answer
C:For the stated requirement to execute a known approval workflow where inputs must be validated before a model drafts a response, explicit workflow stages make ordering and failure handling reproducible; free-form agent planning is unnecessary for a fixed process; this is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is to reduce latency when separate read-only lookups have no ordering dependency, parallelizing independent operations shortens critical-path latency but is unsafe when one action requires another's output; this does not directly meet the required outcome to execute a known approval workflow where inputs must be validated before a model drafts a response.
B:For a system whose objective is to allow a model to request multiple independent data lookups while limiting cost and duration, an orchestration loop can inspect tool requests, execute authorized tools, return results, and terminate under explicit limits; this does not directly meet the required outcome to execute a known approval workflow where inputs must be validated before a model drafts a response.
D:For a system whose objective is to choose between candidate prompt versions by measured task quality rather than intuition, paired evaluation on identical cases isolates prompt impact; production traffic shifts make uncontrolled comparisons misleading; this does not directly meet the required outcome to execute a known approval workflow where inputs must be validated before a model drafts a response.
Question 149
A cloud engineering team needs its AI workflow to prevent malformed model-generated parameters from reaching a write-capable API. What technical approach most directly enables that?
- Use token-based credentials rather than embedded keys
- Use passage-level citations in RAG answers
- Validate tool arguments before invoking actions
- Handle streamed generation with cancellation
Explanation
Correct Answer
C:For the stated requirement to prevent malformed model-generated parameters from reaching a write-capable API, tool schemas guide argument shape, but the application must check types, constraints, identity and authorization before execution; this is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is to authenticate a server-side generative app through Entra where supported, a workload identity and appropriate RBAC reduce static credential exposure while preserving auditable authorization decisions; this does not directly meet the required outcome to prevent malformed model-generated parameters from reaching a write-capable API.
B:For a system whose objective is to let a compliance reviewer locate the specific evidence supporting each generated assertion, citations tied to retrieved chunks improve traceability; a source list without claim alignment does not prove each claim; this does not directly meet the required outcome to prevent malformed model-generated parameters from reaching a write-capable API.
D:For a system whose objective is to show partial assistant output to a user while stopping work after the request is canceled, streaming delivers incremental output; cancellation and backpressure prevent orphaned inference work and stale interface content; this does not directly meet the required outcome to prevent malformed model-generated parameters from reaching a write-capable API.
Question 150
The support ticket describes this behavior: A quote assistant produces inconsistent totals after discounts and tax rules are updated. Which engineering change should come first?
- Retrieve current authorized passages and cite them when forming the response
- Define a typed calculation tool and have the application execute it with validated arguments
- Provide image content to a vision-capable model with a bounded question and verification rubric
- Implement a controlled request-tool-result loop with a maximum number of steps
Explanation
Correct Answer
B:The fault signal is that a quote assistant produces inconsistent totals after discounts and tax rules are updated; the targeted fix works because a tool call delegates deterministic computation to trusted code instead of relying on a model's arithmetic approximation, restoring the ability to obtain exact tax or pricing totals from an authoritative business calculation endpoint.
Incorrect Answers
A:If the failing function were to answer questions about changing internal procedures with source-backed responses, this change could be relevant because RAG retrieves relevant evidence at runtime and gives it to a generative model; model pretraining alone cannot guarantee current policy text; the actual symptom is that a quote assistant produces inconsistent totals after discounts and tax rules are updated, which calls for a different repair.
C:If the failing function were to reason about a photo and related user instructions together in one supported inference flow, this change could be relevant because a multimodal model can interpret visual input alongside textual context; relying on OCR alone loses non-textual relationships; the actual symptom is that a quote assistant produces inconsistent totals after discounts and tax rules are updated, which calls for a different repair.
D:If the failing function were to allow a model to request multiple independent data lookups while limiting cost and duration, this change could be relevant because an orchestration loop can inspect tool requests, execute authorized tools, return results, and terminate under explicit limits; the actual symptom is that a quote assistant produces inconsistent totals after discounts and tax rules are updated, which calls for a different repair.