Question 190
A Foundry-based application must consistently authenticate a server-side generative app through Entra where supported. Which architecture decision is most defensible?
- Validate tool arguments before invoking actions
- Use token-based credentials rather than embedded keys
- Run independent tool requests concurrently where safe
- Resolve deployment and endpoint from runtime configuration
Explanation
Correct Answer
B:For the stated requirement 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 is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is 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 does not directly meet the required outcome to authenticate a server-side generative app through Entra where supported.
C: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 authenticate a server-side generative app through Entra where supported.
D:For a system whose objective is to avoid routing production questions to a test model through hardcoded SDK settings, endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project; this does not directly meet the required outcome to authenticate a server-side generative app through Entra where supported.
Question 191
An engineer receives this incident report: An assistant reads labels correctly but cannot judge how photographed parts are connected. What should be changed?
- Retrieve current authorized passages and cite them when forming the response
- Provide image content to a vision-capable model with a bounded question and verification rubric
- Request structured outputs conforming to the expected schema and validate before ingestion
- Configure a credential chain using the workload identity and least-privilege role grants
Explanation
Correct Answer
B:The fault signal is that an assistant reads labels correctly but cannot judge how photographed parts are connected; the targeted fix works because a multimodal model can interpret visual input alongside textual context; relying on OCR alone loses non-textual relationships, restoring the ability to reason about a photo and related user instructions together in one supported inference flow.
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 an assistant reads labels correctly but cannot judge how photographed parts are connected, which calls for a different repair.
C:If the failing function were to send model-produced classifications into a strict downstream API contract, this change could be relevant because schema-constrained generation helps enforce machine-readable fields and types; application-level validation is still needed for business rules; the actual symptom is that an assistant reads labels correctly but cannot judge how photographed parts are connected, which calls for a different repair.
D: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 an assistant reads labels correctly but cannot judge how photographed parts are connected, which calls for a different repair.
Question 192
The team plans to evaluate answer relevance against user intent. Which test provides the clearest evidence that this change works?
- Compute unsupported-claim rate on source-grounded summaries
- Compare win rate, task accuracy, and safety failures with statistical uncertainty
- Track relevance scores separately from citation-groundedness metrics
- Compare critical-path latency and result consistency with serial execution
Explanation
Correct Answer
C:The accepted change is to evaluate answer relevance against user intent; this evaluation checks whether it can detect answers that are grounded in retrieved text but do not address the actual question, since relevance evaluates how directly a response serves the request; factual grounding and response usefulness are related but distinct axes.
Incorrect Answers
A: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 evaluate answer relevance against user intent works for the current release.
B:This test concerns the ability to choose between candidate prompt versions by measured task quality rather than intuition, for which paired evaluation on identical cases isolates prompt impact; production traffic shifts make uncontrolled comparisons misleading; it does not establish whether the change to evaluate answer relevance against user intent works for the current release.
D:This test concerns the ability to reduce latency when separate read-only lookups have no ordering dependency, for which parallelizing independent operations shortens critical-path latency but is unsafe when one action requires another's output; it does not establish whether the change to evaluate answer relevance against user intent works for the current release.
Question 193
A new Foundry-based application must verify that an app recovers safely from empty messages, long context, and invalid tool responses. Which approach should the team select first?
- Measure factual faithfulness to source context
- Test application robustness on malformed inputs
- Handle streamed generation with cancellation
- Use comparative evaluations for prompt changes
Explanation
Correct Answer
B:For the stated requirement to verify that an app recovers safely from empty messages, long context, and invalid tool responses, boundary-case testing surfaces parsing and orchestration failures that standard benchmark prompts miss; this is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is 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 does not directly meet the required outcome to verify that an app recovers safely from empty messages, long context, and invalid tool responses.
C: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 verify that an app recovers safely from empty messages, long context, and invalid tool responses.
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 verify that an app recovers safely from empty messages, long context, and invalid tool responses.
Question 194
A solution architect is deciding how to detect answers that are grounded in retrieved text but do not address the actual question. Which Azure AI pattern addresses that task?
- Resolve deployment and endpoint from runtime configuration
- Evaluate answer relevance against user intent
- Measure factual faithfulness to source context
- Use comparative evaluations for prompt changes
Explanation
Correct Answer
B:For the stated requirement 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 is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is to avoid routing production questions to a test model through hardcoded SDK settings, endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project; this does not directly meet the required outcome to detect answers that are grounded in retrieved text but do not address the actual question.
C:For a system whose objective is 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 does not directly meet the required outcome to detect answers that are grounded in retrieved text but do not address the actual question.
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 detect answers that are grounded in retrieved text but do not address the actual question.
Question 195
An Azure AI workload passes basic health probes but has this issue: Answers mention a document title but do not reveal which passage supports a specific exemption. Which fix targets it?
- Attach retrieved chunk references to the claims they substantiate and validate that mapping
- Filter authorized indexed chunks using tenant, department, and effective-date fields
- Adjust chunk boundaries and limited overlap, then rebuild affected embeddings
- Validate eligibility first, run generation second, and require approval before sending
Explanation
Correct Answer
A:The fault signal is that answers mention a document title but do not reveal which passage supports a specific exemption; the targeted fix works because citations tied to retrieved chunks improve traceability; a source list without claim alignment does not prove each claim, restoring the ability to let a compliance reviewer locate the specific evidence supporting each generated assertion.
Incorrect Answers
B: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 mention a document title but do not reveal which passage supports a specific exemption, which calls for a different repair.
C:If the failing function were to keep context that crosses page or section boundaries from being lost during indexing, this change could be relevant because chunk overlap can retain nearby context across splits, although excessive overlap inflates index size and duplicates retrieval; the actual symptom is that answers mention a document title but do not reveal which passage supports a specific exemption, which calls for a different repair.
D:If the failing function were to execute a known approval workflow where inputs must be validated before a model drafts a response, this change could be relevant because explicit workflow stages make ordering and failure handling reproducible; free-form agent planning is unnecessary for a fixed process; the actual symptom is that answers mention a document title but do not reveal which passage supports a specific exemption, which calls for a different repair.
Question 196
An engineering team has a requirement to allow a model to request multiple independent data lookups while limiting cost and duration. Which capability should be in the initial design?
- Add bounded tool-use loops for multi-step tasks
- Route conditional branches using rules
- Orchestrate a deterministic sequence around model steps
- Use constrained structured JSON output
Explanation
Correct Answer
A:For the stated requirement 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 is the mechanism that the system must implement.
Incorrect Answers
B:For a system whose objective is to choose among distinct business flows based on a validated account status, rules engines offer predictable decisions for simple structured conditions, while language models handle unstructured interpretation; this does not directly meet the required outcome to allow a model to request multiple independent data lookups while limiting cost and duration.
C:For a system whose objective is 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 does not directly meet the required outcome to allow a model to request multiple independent data lookups while limiting cost and duration.
D:For a system whose objective is to send model-produced classifications into a strict downstream API contract, schema-constrained generation helps enforce machine-readable fields and types; application-level validation is still needed for business rules; this does not directly meet the required outcome to allow a model to request multiple independent data lookups while limiting cost and duration.
Question 197
A cloud engineering team needs its AI workflow to show partial assistant output to a user while stopping work after the request is canceled. What technical approach most directly enables that?
- Resolve deployment and endpoint from runtime configuration
- Validate tool arguments before invoking actions
- Handle streamed generation with cancellation
- Implement fallback when retrieval is insufficient
Explanation
Correct Answer
C:For the stated requirement 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 is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is to avoid routing production questions to a test model through hardcoded SDK settings, endpoint and deployment identifiers are environment-specific; runtime configuration keeps inference calls aligned to the project; this does not directly meet the required outcome to show partial assistant output to a user while stopping work after the request is canceled.
B:For a system whose objective is 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 does not directly meet the required outcome to show partial assistant output to a user while stopping work after the request is canceled.
D:For a system whose objective is to avoid fabricating an answer to a question not covered by the organization's knowledge collection, an evidence threshold and abstention path distinguish unknown facts from model completion skill; confident text is not proof of source support; this does not directly meet the required outcome to show partial assistant output to a user while stopping work after the request is canceled.
Question 198
Production telemetry reveals this failure: An LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag. Which change most directly corrects it?
- Execute independent authorized reads in parallel and join results before generation
- Implement a controlled request-tool-result loop with a maximum number of steps
- Use deterministic status rules for routing and reserve the LLM for language-specific subtasks
- Run A/B offline comparisons on a representative fixed dataset before rollout
- Attach retrieved chunk references to the claims they substantiate and validate that mapping
Explanation
Correct Answer
C:The fault signal is that an LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag; the targeted fix works because rules engines offer predictable decisions for simple structured conditions, while language models handle unstructured interpretation, restoring the ability to choose among distinct business flows based on a validated account status.
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 LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag, which calls for a different repair.
B: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 an LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag, which calls for a different repair.
D:If the failing function were to choose between candidate prompt versions by measured task quality rather than intuition, this change could be relevant because paired evaluation on identical cases isolates prompt impact; production traffic shifts make uncontrolled comparisons misleading; the actual symptom is that an LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag, which calls for a different repair.
E:If the failing function were to let a compliance reviewer locate the specific evidence supporting each generated assertion, this change could be relevant because citations tied to retrieved chunks improve traceability; a source list without claim alignment does not prove each claim; the actual symptom is that an LLM occasionally sends blocked accounts through a fulfillment branch despite a clear status flag, which calls for a different repair.
Question 199
What test should engineers run after they apply retrieval metadata filters before generation, rather than relying solely on a successful API response?
- Test for cross-department leakage and expired-document citation rate
- Check claim-to-citation accuracy on questions with several similar policy sections
- Track relevance scores separately from citation-groundedness metrics
- Test retrieval of examples whose supporting clauses straddle chunk boundaries
Explanation
Correct Answer
A:The accepted change is to apply retrieval metadata filters before generation; this evaluation checks whether it can ensure answers include only documents for a user's department and effective policy year, since retrieval filters narrow the evidence set by structured attributes before the model sees it; prompt instructions are not security filters.
Incorrect Answers
B:This test concerns the ability to let a compliance reviewer locate the specific evidence supporting each generated assertion, for which citations tied to retrieved chunks improve traceability; a source list without claim alignment does not prove each claim; it does not establish whether the change to apply retrieval metadata filters before generation works for the current release.
C:This test concerns the ability to detect answers that are grounded in retrieved text but do not address the actual question, for which relevance evaluates how directly a response serves the request; factual grounding and response usefulness are related but distinct axes; it does not establish whether the change to apply retrieval metadata filters before generation works for the current release.
D:This test concerns the ability to keep context that crosses page or section boundaries from being lost during indexing, for which chunk overlap can retain nearby context across splits, although excessive overlap inflates index size and duplicates retrieval; it does not establish whether the change to apply retrieval metadata filters before generation works for the current release.
Question 200
In an architecture review, the key requirement is to reduce latency when separate read-only lookups have no ordering dependency. Which solution directly satisfies it?
- Orchestrate a deterministic sequence around model steps
- Handle streamed generation with cancellation
- Run independent tool requests concurrently where safe
- Route conditional branches using rules
Explanation
Correct Answer
C:For the stated requirement 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 is the mechanism that the system must implement.
Incorrect Answers
A:For a system whose objective is 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 does not directly meet the required outcome to reduce latency when separate read-only lookups have no ordering dependency.
B: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 reduce latency when separate read-only lookups have no ordering dependency.
D:For a system whose objective is to choose among distinct business flows based on a validated account status, rules engines offer predictable decisions for simple structured conditions, while language models handle unstructured interpretation; this does not directly meet the required outcome to reduce latency when separate read-only lookups have no ordering dependency.