Question 180
The team reduces completion tokens but still receives context-window errors on very large prompts. Which limit was confused?
- Inspect the selected model's supported sampling parameters and omit unsupported ones
- Separate example cases from held-out evaluation questions
- Declare all required fields in the supported structured-output schema
- Distinguish input context-window constraints from output-token limits
Explanation
Correct Answer
D:Limiting generation length does not necessarily make oversized prompt input fit a model's context window. In the instruction-following chat assistant, confirm the fix by retesting the affected behavior after distinguishing input context-window constraints from output-token limits.
Incorrect Answers
A:Although useful elsewhere in the instruction-following chat assistant, generation controls vary by model and API; a universal temperature setting is not valid for all deployments; the present case is different and calls for the team to distinguish input context-window constraints from output-token limits.
B:Leaking test examples into prompt demonstrations inflates measured quality and fails to test generalization in its own workflow, but that is a different correction from the present requirement to distinguish input context-window constraints from output-token limits. The reported issue would remain.
C:That change addresses another failure mode: jSON validity alone does not ensure completeness; required properties define the application's contract; it does not address this case, for which the relevant intervention is to distinguish input context-window constraints from output-token limits.
Question 181
A team responsible for a team-developed knowledge assistant needs to call project resources and supported model or agent operations in application code; the acceptance check is that the application's code invokes the intended Foundry project operations. Which capability should be selected?
- Integrate the application through the Foundry SDK
- Use a Foundry project workspace
- Run a model or flow evaluation
- Build a prompt flow
Explanation
Correct Answer
A:The SDK provides supported programmatic interfaces for the selected Foundry environment and API surface. For this team-developed knowledge assistant, the intended evidence is that the application's code invokes the intended Foundry project operations. Installing an SDK does not eliminate the need for authentication and deployed resources.
Incorrect Answers
B:A project provides the working context for developing AI applications and managing associated resources; it may be used to keep the application's resources, connections, and evaluation assets organized for developers in a team-developed knowledge assistant, but the current requirement is instead to call project resources and supported model or agent operations in application code. That separate capability does not prove that the application's code invokes the intended Foundry project operations; The project itself is not a text embedding model or search query, so selecting that alternative would not meet the stated acceptance check.
C:Evaluation applies metrics or evaluators to test responses, enabling systematic comparison of changes; it may be used to compare response quality and safety on representative inputs against specified criteria in a team-developed knowledge assistant, but the current requirement is instead to call project resources and supported model or agent operations in application code. That separate capability does not prove that the application's code invokes the intended Foundry project operations; It is distinct from publishing a model endpoint or buying throughput capacity, so selecting that alternative would not meet the stated acceptance check.
D:Prompt flow represents a connected sequence of executable steps that can be tested and compared; it may be used to connect prompt, model, and processing nodes in a reproducible workflow in a team-developed knowledge assistant, but the current requirement is instead to call project resources and supported model or agent operations in application code. That separate capability does not prove that the application's code invokes the intended Foundry project operations; It does not automatically make all enterprise documents available to a model, so selecting that alternative would not meet the stated acceptance check.
Question 182
A new model looks good on the average benchmark score, but safety reviewers must decide whether to release it for sensitive client use. Which THREE reviews should complement the average score? (Choose THREE.)
- Measure performance across relevant user and task subgroups
- Associate each application's evaluation datasets and runs with its intended project
- Create the named deployment and use its exact identifier in inference requests
- Review safety and policy-adherence results on representative high-risk tests
- Inspect individual failed examples and error patterns
- Use a deployed workload identity with supported role assignments instead of interactive login
Explanation
Correct Answers
A:Subgroup analysis can expose concentrated failures on languages, intent types, or business contexts that are uncommon in aggregate results.
D:A fluent answer may still violate safety requirements, so the release decision needs dedicated assessments of harmful and disallowed behavior.
E:Examples of incorrect, misleading, or ungrounded responses reveal failure modes that aggregate scores can hide from reviewers.
Incorrect Answers
B:Although useful elsewhere in the team-developed knowledge assistant, separating project assets avoids mixing comparisons and experiment history across products; the present case is different and calls for the team to inspect individual failed examples and error patterns and measure performance across relevant user and task subgroups and review safety and policy-adherence results on representative high-risk tests.
C:That change addresses another failure mode: catalog listings describe available models, while inference addresses actual deployed instances; it does not address this case, for which the relevant intervention is to inspect individual failed examples and error patterns and measure performance across relevant user and task subgroups and review safety and policy-adherence results on representative high-risk tests.
F:Production automation should authenticate as a governed principal rather than relying on a developer's session in its own workflow, but that is a different correction from the present requirement to inspect individual failed examples and error patterns and measure performance across relevant user and task subgroups and review safety and policy-adherence results on representative high-risk tests. The reported issue would remain.
Question 183
A release for a team-developed knowledge assistant is blocked until the application can keep the application's resources, connections, and evaluation assets organized for developers. The tester will verify that collaborators can locate the same application's managed assets. What should be implemented?
- Use a Foundry project workspace
- Run a model or flow evaluation
- Build a prompt flow
- Integrate the application through the Foundry SDK
Explanation
Correct Answer
A:A project provides the working context for developing AI applications and managing associated resources. For this team-developed knowledge assistant, the intended evidence is that collaborators can locate the same application's managed assets. The project itself is not a text embedding model or search query.
Incorrect Answers
B:Evaluation applies metrics or evaluators to test responses, enabling systematic comparison of changes; it may be used to compare response quality and safety on representative inputs against specified criteria in a team-developed knowledge assistant, but the current requirement is instead to keep the application's resources, connections, and evaluation assets organized for developers. That separate capability does not prove that collaborators can locate the same application's managed assets; It is distinct from publishing a model endpoint or buying throughput capacity, so selecting that alternative would not meet the stated acceptance check.
C:Prompt flow represents a connected sequence of executable steps that can be tested and compared; it may be used to connect prompt, model, and processing nodes in a reproducible workflow in a team-developed knowledge assistant, but the current requirement is instead to keep the application's resources, connections, and evaluation assets organized for developers. That separate capability does not prove that collaborators can locate the same application's managed assets; It does not automatically make all enterprise documents available to a model, so selecting that alternative would not meet the stated acceptance check.
D:The SDK provides supported programmatic interfaces for the selected Foundry environment and API surface; it may be used to call project resources and supported model or agent operations in application code in a team-developed knowledge assistant, but the current requirement is instead to keep the application's resources, connections, and evaluation assets organized for developers. That separate capability does not prove that collaborators can locate the same application's managed assets; Installing an SDK does not eliminate the need for authentication and deployed resources, so selecting that alternative would not meet the stated acceptance check.
Question 184
A CI pipeline fails after the SDK interface changes across machines. What deployment practice reduces drift?
- Target the intended production deployment name when running release validation
- Pin compatible SDK dependency versions and validate upgrades in staging
- Parameterize the prompt-flow inputs and run both variants on a shared test dataset
- Freeze and version the evaluation dataset and scoring criteria for regression runs
Explanation
Correct Answer
B:Reproducible dependency versions prevent silent breaking changes during releases. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after pinning compatible SDK dependency versions and validate upgrades in staging.
Incorrect Answers
A:Test results are meaningful only for the deployment clients will actually call in its own workflow, but that is a different correction from the present requirement to pin compatible SDK dependency versions and validate upgrades in staging. The reported issue would remain.
C:That change addresses another failure mode: repeatable parameterized flow runs support controlled comparisons rather than ad hoc edits; it does not address this case, for which the relevant intervention is to pin compatible SDK dependency versions and validate upgrades in staging.
D:Although useful elsewhere in the team-developed knowledge assistant, controlled datasets and criteria allow meaningful before/after comparisons; the present case is different and calls for the team to pin compatible SDK dependency versions and validate upgrades in staging.
Question 185
A team wants to compare two prompt variations without changing application source code each time. What workflow practice is useful?
- Freeze and version the evaluation dataset and scoring criteria for regression runs
- Pin compatible SDK dependency versions and validate upgrades in staging
- Parameterize the prompt-flow inputs and run both variants on a shared test dataset
- Target the intended production deployment name when running release validation
Explanation
Correct Answer
C:Repeatable parameterized flow runs support controlled comparisons rather than ad hoc edits. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after parameterizing the prompt-flow inputs and run both variants on a shared test dataset.
Incorrect Answers
A:That change addresses another failure mode: controlled datasets and criteria allow meaningful before/after comparisons; it does not address this case, for which the relevant intervention is to parameterize the prompt-flow inputs and run both variants on a shared test dataset.
B:Reproducible dependency versions prevent silent breaking changes during releases in its own workflow, but that is a different correction from the present requirement to parameterize the prompt-flow inputs and run both variants on a shared test dataset. The reported issue would remain.
D:Although useful elsewhere in the team-developed knowledge assistant, test results are meaningful only for the deployment clients will actually call; the present case is different and calls for the team to parameterize the prompt-flow inputs and run both variants on a shared test dataset.
Question 186
A team-developed knowledge assistant must connect prompt, model, and processing nodes in a reproducible workflow. Reviewers will accept the implementation when a defined flow run produces intermediate and final node outputs. Which approach best fulfills that requirement?
- Run a model or flow evaluation
- Build a prompt flow
- Integrate the application through the Foundry SDK
- Deploy an appropriate model from the catalog
Explanation
Correct Answer
B:Prompt flow represents a connected sequence of executable steps that can be tested and compared. For this team-developed knowledge assistant, the intended evidence is that a defined flow run produces intermediate and final node outputs. It does not automatically make all enterprise documents available to a model.
Incorrect Answers
A:Evaluation applies metrics or evaluators to test responses, enabling systematic comparison of changes; it may be used to compare response quality and safety on representative inputs against specified criteria in a team-developed knowledge assistant, but the current requirement is instead to connect prompt, model, and processing nodes in a reproducible workflow. That separate capability does not prove that a defined flow run produces intermediate and final node outputs; It is distinct from publishing a model endpoint or buying throughput capacity, so selecting that alternative would not meet the stated acceptance check.
C:The SDK provides supported programmatic interfaces for the selected Foundry environment and API surface; it may be used to call project resources and supported model or agent operations in application code in a team-developed knowledge assistant, but the current requirement is instead to connect prompt, model, and processing nodes in a reproducible workflow. That separate capability does not prove that a defined flow run produces intermediate and final node outputs; Installing an SDK does not eliminate the need for authentication and deployed resources, so selecting that alternative would not meet the stated acceptance check.
D:A model deployment provides an inference target with a deployment name and the selected model's capabilities; it may be used to make a selected compatible generative model callable by application code in a team-developed knowledge assistant, but the current requirement is instead to connect prompt, model, and processing nodes in a reproducible workflow. That separate capability does not prove that a defined flow run produces intermediate and final node outputs; Selecting a catalog entry alone does not produce a callable deployment, so selecting that alternative would not meet the stated acceptance check.
Question 187
A flow produces a final response but the developer cannot tell which tool node supplied an incorrect value. What should the workflow expose?
- Capture intermediate node inputs and outputs for the flow run
- Score both response quality and groundedness on the same held-out examples
- Use a model deployment documented to support the requested image-input modality
- Use governed project connections and shared evaluation assets rather than private local settings
Explanation
Correct Answer
A:Node-level execution results make data movement and failures observable across multi-step prompt processing. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after capturing intermediate node inputs and outputs for the flow run.
Incorrect Answers
B:That change addresses another failure mode: single-metric fluency assessments miss whether a response faithfully uses retrieved evidence; it does not address this case, for which the relevant intervention is to capture intermediate node inputs and outputs for the flow run.
C:Although useful elsewhere in the team-developed knowledge assistant, endpoint availability does not imply every model accepts every input type; the present case is different and calls for the team to capture intermediate node inputs and outputs for the flow run.
D:Project-managed artifacts let collaborators use a consistent application environment while retaining access controls in its own workflow, but that is a different correction from the present requirement to capture intermediate node inputs and outputs for the flow run. The reported issue would remain.
Question 188
A new prompt increases fluency but worsens citation fidelity on internal FAQs. What evaluation design will detect the trade-off?
- Score both response quality and groundedness on the same held-out examples
- Use a model deployment documented to support the requested image-input modality
- Load both the intended Foundry project connection and the target model deployment
- Capture intermediate node inputs and outputs for the flow run
Explanation
Correct Answer
A:Single-metric fluency assessments miss whether a response faithfully uses retrieved evidence. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after scoring both response quality and groundedness on the same held-out examples.
Incorrect Answers
B:Although useful elsewhere in the team-developed knowledge assistant, endpoint availability does not imply every model accepts every input type; the present case is different and calls for the team to score both response quality and groundedness on the same held-out examples.
C:That change addresses another failure mode: sDK clients still need concrete inference targets; a project endpoint alone does not name the model to serve a request; it does not address this case, for which the relevant intervention is to score both response quality and groundedness on the same held-out examples.
D:Node-level execution results make data movement and failures observable across multi-step prompt processing in its own workflow, but that is a different correction from the present requirement to score both response quality and groundedness on the same held-out examples. The reported issue would remain.
Question 189
A call succeeds for a text-only deployment but fails when the app includes an image. What should the deployment selection verify?
- Capture intermediate node inputs and outputs for the flow run
- Use governed project connections and shared evaluation assets rather than private local settings
- Use a model deployment documented to support the requested image-input modality
- Load both the intended Foundry project connection and the target model deployment
Explanation
Correct Answer
C:Endpoint availability does not imply every model accepts every input type. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after using a model deployment documented to support the requested image-input modality.
Incorrect Answers
A:Node-level execution results make data movement and failures observable across multi-step prompt processing in its own workflow, but that is a different correction from the present requirement to use a model deployment documented to support the requested image-input modality. The reported issue would remain.
B:That change addresses another failure mode: project-managed artifacts let collaborators use a consistent application environment while retaining access controls; it does not address this case, for which the relevant intervention is to use a model deployment documented to support the requested image-input modality.
D:Although useful elsewhere in the team-developed knowledge assistant, sDK clients still need concrete inference targets; a project endpoint alone does not name the model to serve a request; the present case is different and calls for the team to use a model deployment documented to support the requested image-input modality.
Question 190
An app uses developer credentials locally but must run unattended in Azure production. What authentication change is preferred?
- Use a deployed workload identity with supported role assignments instead of interactive login
- Create the named deployment and use its exact identifier in inference requests
- Use a secured project connection or approved secret injection for the external call
- Examine failure examples, subgroup performance, and safety results alongside aggregates
Explanation
Correct Answer
A:Production automation should authenticate as a governed principal rather than relying on a developer's session. In the team-developed knowledge assistant, confirm the fix by retesting the affected behavior after using a deployed workload identity with supported role assignments instead of interactive login.
Incorrect Answers
B:Although useful elsewhere in the team-developed knowledge assistant, catalog listings describe available models, while inference addresses actual deployed instances; the present case is different and calls for the team to use a deployed workload identity with supported role assignments instead of interactive login.
C:That change addresses another failure mode: flows should resolve managed credentials; embedding secrets in workflow definitions risks exposure; it does not address this case, for which the relevant intervention is to use a deployed workload identity with supported role assignments instead of interactive login.
D:Mean scores can conceal severe errors on rare or high-impact cases in its own workflow, but that is a different correction from the present requirement to use a deployed workload identity with supported role assignments instead of interactive login. The reported issue would remain.