Microsoft AI-103 Practice Test 32
Question 310
An indexing job finishes without errors but fewer documents are searchable than in the source system. What monitoring is needed?
- Declare ingestion complete because the indexing job returned HTTP 200, without reconciling rejected or missing document IDs.
- Run the same labeled retrieval benchmark against old and candidate index configurations
- Compare expected source counts with indexed and rejected item counts
- Track indexing load and query latency and throttling together
Explanation
Correct Answer
C:Coverage checks reveal silent omissions not captured by a generic success status. The decisive difference is whether the design can detect ingestion completeness problems; the competing function serves to isolate which ingestion stage fails.
Incorrect Answers
A:A successful job response does not prove every source item reached the search index. Reconcile source counts and IDs against indexed, skipped and failed records to detect gaps.
B:Controlled comparisons distinguish relevance changes from differing query distributions. Here the needed outcome is to detect ingestion completeness problems; an option designed to measure retrieval regressions across index changes would not directly provide it.
D:Ingestion workload can compete with query capacity and affect interactive performance. That is relevant to detect index refresh effects on search service, but the present scenario requires an ability to detect ingestion completeness problems.
Question 311
A privacy analysis cannot begin until a legal agent identifies applicable jurisdiction, but pricing can run independently. How should tasks be scheduled?
- Use a coordinating agent to assign specialists and consolidate sourced results
- Give one owner write authority or serialize merges through the coordinator
- Launch the jurisdiction-dependent privacy assessment before the legal agent identifies the governing jurisdiction.
- Model prerequisite dependencies and parallelize only independent tasks
Explanation
Correct Answer
D:Legal jurisdiction is required input to privacy assessment while pricing has no such dependency. This directly supports the need to run dependent steps in order and unrelated work in parallel, unlike a mechanism primarily intended to coordinate specialists and synthesize results.
Incorrect Answers
A:A central coordinator preserves coverage and resolves conflicts before one answer is returned. The function serves to coordinate specialists and synthesize results, whereas the actual requirement is to run dependent steps in order and unrelated work in parallel.
B:Concurrent stateful writes need ownership and a merge discipline to prevent lost changes. That capability is appropriate for avoid conflicting simultaneous agent edits, not this separate task of finding how to run dependent steps in order and unrelated work in parallel.
C:The privacy analysis depends on legal scope. Run independent pricing in parallel but wait for the jurisdiction prerequisite before privacy review.
Question 312
A user reports newly deleted compliance policies still appear as relevant hits. What index-maintenance behavior is missing?
- Track search query latency, throttling and indexing resource saturation
- Propagate source deletions and tombstones into the search index
- Compare discovered source item counts and indexed document counts by connector
- Run more OCR confidence sampling instead of removing policy documents that were deleted from the authoritative source.
Explanation
Correct Answer
B:Documents removed from the source must be removed or deactivated in retrieval. The decisive difference is whether the design can remove obsolete indexed evidence; the competing function serves to diagnose search infrastructure performance.
Incorrect Answers
A:Index/search capacity and query complexity can cause tail latency independent of model generation. That capability is appropriate for diagnose search infrastructure performance, not this separate task of finding how to remove obsolete indexed evidence.
C:Index job success does not ensure all expected records were ingested; coverage counters reveal omissions. Its primary purpose is to detect missing source records in ingestion; the stated constraint instead demands a way to remove obsolete indexed evidence.
D:Better extraction quality does not remove stale index entries. Reconcile source deletes and invalidate corresponding indexed chunks before retrieval.
Question 313
An analyzer returns a bounding region but an application assumes all coordinates refer to a different resized image. What should be verified?
- Draw the returned bounding coordinates on a resized preview without mapping them from the analyzer's original image geometry.
- Use a single-task Content Understanding analyzer configured for the required fields
- Normalize coordinate systems and image dimensions before overlaying extracted regions
- Route low-confidence extracted fields to review before irreversible decisions
Explanation
Correct Answer
C:Region predictions must be transformed using the source media geometry to avoid wrong annotations. The practical requirement is to align extracted regions with rendered images; a solution built to detect missed visual labels quantitatively would not meet that same need.
Incorrect Answers
A:Untransformed bounding boxes can be shifted or scaled incorrectly. Verify the coordinate reference, crop and dimensions before overlaying regions.
B:A focused extraction task can use a single-task pipeline without broad open-ended analysis. This helps when a team needs to perform one narrowly scoped extraction task; it does not establish how to align extracted regions with rendered images.
D:Confidence and validation checks reduce high-impact errors from uncertain extraction. That capability is appropriate for escalate uncertain extracted visual fields, not this separate task of finding how to align extracted regions with rendered images.
Question 314
An accessibility app speaks product names so fast they are unintelligible. What output parameter should be adjusted?
- Accelerate speech synthesis further to reduce the duration of an already unintelligible reply
- Control the supported speech rate and pronunciation settings
- Choose a compatible language and voice configuration for synthesis
- Use text-to-speech synthesis for the response text
Explanation
Correct Answer
B:Speech rendering controls can improve intelligibility without altering the underlying facts. The practical requirement is to improve spoken output pacing; a solution built to manage storage and privacy for synthesized replies would not meet that same need.
Incorrect Answers
A:Faster output would worsen comprehension. Adjust pace and pronunciation and test with representative listeners.
C:An appropriate voice/language match improves pronunciation and user experience. The function serves to synthesize speech in the selected language, whereas the actual requirement is to improve spoken output pacing.
D:Speech synthesis converts generated or predetermined text into audio suitable for playback. That capability is appropriate for turn agent text into spoken output, not this separate task of finding how to improve spoken output pacing.
Question 315
A pipeline has high error rates after receiving PDFs with a new layout. Which metric identifies the failing stage?
- Track extraction/OCR and enrichment errors separately from indexing success
- Continue parsing the changed PDF format using unverified old coordinates
- Run the same labeled retrieval benchmark against old and candidate index configurations
- Compare expected source counts with indexed and rejected item counts
Explanation
Correct Answer
A:Stage-specific health shows whether the problem lies in parsing, enrichment or index writes. In this scenario the team must isolate which ingestion stage fails. A feature used to detect index refresh effects on search service does not substitute for it.
Incorrect Answers
B:A layout change can shift fields while still producing plausible strings. Extraction should be evaluated and adapted to the new document structure.
C:Controlled comparisons distinguish relevance changes from differing query distributions. That is relevant to measure retrieval regressions across index changes, but the present scenario requires an ability to isolate which ingestion stage fails.
D:Coverage checks reveal silent omissions not captured by a generic success status. This helps when a team needs to detect ingestion completeness problems; it does not establish how to isolate which ingestion stage fails.
Question 316
Two coding agents edit the same configuration file concurrently and overwrite each other. What concurrency strategy helps?
- Let all coding agents overwrite the shared configuration concurrently and accept the last write as authoritative.
- Dispatch only the capabilities needed by the request
- Give one owner write authority or serialize merges through the coordinator
- Require source references, scope and confidence in each specialist deliverable
Explanation
Correct Answer
C:Concurrent stateful writes need ownership and a merge discipline to prevent lost changes. The decisive difference is whether the design can avoid conflicting simultaneous agent edits; the competing function serves to reconcile contradictory specialist claims.
Incorrect Answers
A:Last-write-wins can silently discard correct concurrent changes. Give one owner write authority or serialize reviewed merges.
B:Selective orchestration avoids unnecessary model calls, latency and possible conflicting outputs. Here the needed outcome is to avoid conflicting simultaneous agent edits; an option designed to avoid unnecessary agent fan-out would not directly provide it.
D:Traceable outputs let the coordinating agent verify and cite claims instead of merging unsupported summaries. That is relevant to make specialist results auditable, but the present scenario requires an ability to avoid conflicting simultaneous agent edits.
Question 317
A service handles private employee records and must use an authorized tool on behalf of the caller. What boundary prevents cross-user leakage?
- Authorize based solely on the agent service identity's broad resource role.
- Enforce caller-specific identity and document authorization at every retrieval and tool access.
- Filter unauthorized employee names out of the final answer after fetching records.
- Use one shared cache of retrieved personnel data across all callers.
Explanation
Correct Answer
B:An agent's service identity does not permit broad cross-tenant access. Authentication and per-record authorization must be checked for the requesting user.
Incorrect Answers
A:A service principal's technical reach does not equal the invoking user's permission to private records.
C:Post-output filtering cannot undo disclosure to the model during retrieval or tool execution.
D:A cross-user cache can return material outside an individual's approved access scope.
Question 318
A scanned form contains a table whose columns are swapped by the downstream parser. What extraction output should be preserved?
- Flatten all recognized table words into one untyped sentence and discard their row and column associations.
- Evaluate pro-mode Content Understanding for a multi-stage analysis flow
- Configure a visual analyzer to identify relevant objects and regions
- Retain layout and table-cell relationships rather than flattening all OCR words
Explanation
Correct Answer
D:Structured layout informs which values belong to which headings and rows. Here the relevant capability must let the team preserve table structure from document images. An alternative targeting detect missed visual labels quantitatively solves another problem.
Incorrect Answers
A:The application must preserve cell-to-column relationships. Flattened OCR text cannot reliably distinguish each amount or its labeled column.
B:Advanced workflows may need a more capable orchestration mode than one fixed extraction. The function serves to run a richer multimodal understanding pipeline, whereas the actual requirement is to preserve table structure from document images.
C:Visual extraction can return object-level evidence for downstream validation or action. That capability is appropriate for extract object and region information from an image, not this separate task of finding how to preserve table structure from document images.
Question 319
A speech recognizer misattributes one customer's sentence to an agent during a recorded support call. What feature is missing?
- Improve punctuation in the final transcript but ignore speaker boundaries.
- Enable and evaluate speaker diarization with the transcript's timestamps.
- Enable text-to-speech voice style transfer on the recording.
- Remove timestamps and combine all recognized text into one paragraph.
Explanation
Correct Answer
B:Diarization maps recognized speech segments to speakers; word accuracy alone cannot identify who said each sentence.
Incorrect Answers
A:Better punctuation does not label which participant produced each utterance.
C:Speech synthesis controls output voices and does not identify input speakers.
D:Flattening the timeline destroys evidence needed to assign turns to speakers.
Question 320
A Foundry Content Understanding analyzer emits clean Markdown, but the business app needs typed invoice fields. What should be configured?
- Use a plain Markdown output without any field definitions and parse it heuristically.
- Increase the OCR resolution while retaining untyped narrative-only output.
- Ask the agent to guess column names from whichever invoice arrives next.
- Select or define an analyzer output schema that includes the required structured fields.
Explanation
Correct Answer
D:Markdown captures readable content; a downstream accounting interface may require a validated field schema.
Incorrect Answers
A:Markdown layout may be readable but does not guarantee typed invoice fields for the application.
B:Better character recognition does not define a structured output schema for fields.
C:Uncontrolled naming produces unstable field contracts and breaks downstream processing.