100% Latest & Updated Microsoft DP-800 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 790
A vector pipeline periodically changes model from version A to B. The team selects: Version embeddings and re-embed documents consistently. Which implementation action puts this design into practice?
- Normalize vectors at ingestion and query where required
- Persist source ID, chunk offset, model version and timestamp
- Store embedding_model_version per row and build replacement index
- Capture committed row changes and queue changed document IDs
Explanation
Correct Answer
C:Store embedding_model_version per row and build replacement index. This is the concrete setup for version embeddings and re-embed documents consistently, and query only a single compatible model version in test supplies a direct confirmation.
Incorrect Answers
A:Normalize vectors at ingestion and query where required. This concerns consistent normalization. Instead, store embedding_model_version per row and build replacement index.
B:Persist source ID, chunk offset, model version and timestamp. This concerns an app stores user questions next to vector data. Instead, store embedding_model_version per row and build replacement index.
D:Capture committed row changes and queue changed document IDs. This concerns a committed row-change feed with before/after column values for a downstream embedding service. Instead, store embedding_model_version per row and build replacement index.
Question 791
A vector pipeline periodically changes model from version A to B. After implementing the design, which check most directly validates the expected behavior?
- Test token acquisition and approved inference call
- Query only a single compatible model version in test
- Run job twice and confirm no duplicate chunks
- Compare outputs to source IDs and model version
Explanation
Correct Answer
B:Query only a single compatible model version in test. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that old and new vectors were mixed in one ranking space.
Incorrect Answers
A:Test token acquisition and approved inference call. This concerns a SQL ai solution consumes an external model endpoint. Instead, query only a single compatible model version in test.
C:Run job twice and confirm no duplicate chunks. This concerns being replayable after a crash. Instead, query only a single compatible model version in test.
D:Compare outputs to source IDs and model version. This concerns being embedded efficiently. Instead, query only a single compatible model version in test.
Question 792
A vector pipeline periodically changes model from version A to B. A test instead finds: Results deteriorate after partial migration. Which underlying issue best explains this finding?
- Embedding dimension differs from VECTOR column definition
- CDC consumer checkpoint is stalled
- Pipeline generates new random chunk IDs each run
- Old and new vectors were mixed in one ranking space
Explanation
Correct Answer
D:Old and new vectors were mixed in one ranking space. It accounts for the specific observed failure: results deteriorate after partial migration. The intended solution is version embeddings and re-embed documents consistently with the required setup.
Incorrect Answers
A:Embedding dimension differs from VECTOR column definition. This concerns database rejects new vectors as incompatible. Here, old and new vectors were mixed in one ranking space.
B:CDC consumer checkpoint is stalled. This concerns vectors lag behind live catalog data. Here, old and new vectors were mixed in one ranking space.
C:Pipeline generates new random chunk IDs each run. This concerns rerun doubles search documents. Here, old and new vectors were mixed in one ranking space.
Question 793
A search system must regenerate embeddings after source documents are deleted. Which database design or engineering approach most directly meets this requirement?
- Version embeddings and re-embed documents consistently
- Delete tombstoned vector rows on source removal
- Evaluate data residency and provider handling requirements
- Keep source text and provenance IDs with embeddings
Explanation
Correct Answer
B:Delete tombstoned vector rows on source removal. The implementation is process deletion events and purge embeddings by source id; success can be checked by delete a document and confirm no vector result remains. The method addresses the stated data and workload constraint.
Incorrect Answers
A:Version embeddings and re-embed documents consistently. This concerns a vector pipeline periodically changes model from version a to b. Here, delete tombstoned vector rows on source removal.
C:Evaluate data residency and provider handling requirements. This concerns a privacy-sensitive organization connects SQL to an external model. Here, delete tombstoned vector rows on source removal.
D:Keep source text and provenance IDs with embeddings. This concerns an app stores user questions next to vector data. Here, delete tombstoned vector rows on source removal.
Question 794
A search system must regenerate embeddings after source documents are deleted. The team selects: Delete tombstoned vector rows on source removal. Which implementation action puts this design into practice?
- Apply approved column selection and redaction pipeline
- Process deletion events and purge embeddings by source ID
- Persist source ID, chunk offset, model version and timestamp
- Store model ID and run same model for queries and documents
Explanation
Correct Answer
B:Process deletion events and purge embeddings by source ID. This is the concrete setup for delete tombstoned vector rows on source removal, and delete a document and confirm no vector result remains supplies a direct confirmation.
Incorrect Answers
A:Apply approved column selection and redaction pipeline. This concerns not index private identifiers. Instead, process deletion events and purge embeddings by source ID.
C:Persist source ID, chunk offset, model version and timestamp. This concerns an app stores user questions next to vector data. Instead, process deletion events and purge embeddings by source ID.
D:Store model ID and run same model for queries and documents. This concerns cross-language semantic search. Instead, process deletion events and purge embeddings by source ID.
Question 795
A search system must regenerate embeddings after source documents are deleted. After implementing the design, which check most directly validates the expected behavior?
- Retrieve a result and trace it to exact source section
- Update one row and confirm only affected vector refreshes
- Compare outputs to source IDs and model version
- Delete a document and confirm no vector result remains
Explanation
Correct Answer
D:Delete a document and confirm no vector result remains. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that deletion events were ignored by vector maintenance.
Incorrect Answers
A:Retrieve a result and trace it to exact source section. This concerns an app stores user questions next to vector data. Here, delete a document and confirm no vector result remains.
B:Update one row and confirm only affected vector refreshes. This concerns only changed primary keys, not before/after row images. Here, delete a document and confirm no vector result remains.
C:Compare outputs to source IDs and model version. This concerns being embedded efficiently. Here, delete a document and confirm no vector result remains.
Question 796
A search system must regenerate embeddings after source documents are deleted. A test instead finds: Search returns content from removed source. Which underlying issue best explains this finding?
- Vector rows lack stable provenance metadata
- Model selection ignored operational latency constraints
- Embedding dimension differs from VECTOR column definition
- Deletion events were ignored by vector maintenance
Explanation
Correct Answer
D:Deletion events were ignored by vector maintenance. It accounts for the specific observed failure: search returns content from removed source. The intended solution is delete tombstoned vector rows on source removal with the required setup.
Incorrect Answers
A:Vector rows lack stable provenance metadata. This concerns search result cannot be cited back to document. Here, deletion events were ignored by vector maintenance.
B:Model selection ignored operational latency constraints. This concerns huge model exceeds time budget without quality gain. Here, deletion events were ignored by vector maintenance.
C:Embedding dimension differs from VECTOR column definition. This concerns database rejects new vectors as incompatible. Here, deletion events were ignored by vector maintenance.
Question 797
A database contains sections and paragraphs as potential chunks. Which database design or engineering approach most directly meets this requirement?
- Use Change Tracking and an incremental embedding worker
- Choose a model with structured output support
- Choose chunk granularity based on retrieval questions
- Keep source text and provenance IDs with embeddings
Explanation
Correct Answer
C:Choose chunk granularity based on retrieval questions. The implementation is benchmark paragraph and section chunks against evaluation set; success can be checked by measure answer precision and source coverage. The method addresses the stated data and workload constraint.
Incorrect Answers
A:Use Change Tracking and an incremental embedding worker. This concerns only changed primary keys, not before/after row images. Instead, choose chunk granularity based on retrieval questions.
B:Choose a model with structured output support. This concerns deterministic json fields in model output. Instead, choose chunk granularity based on retrieval questions.
D:Keep source text and provenance IDs with embeddings. This concerns an app stores user questions next to vector data. Instead, choose chunk granularity based on retrieval questions.
Question 798
A database contains sections and paragraphs as potential chunks. The team selects: Choose chunk granularity based on retrieval questions. Which implementation action puts this design into practice?
- Call approved embedding model and store VECTOR with matching dimension
- Benchmark paragraph and section chunks against evaluation set
- Configure external model with approved identity and secure URL
- Benchmark representative prompts on candidate models
Explanation
Correct Answer
B:Benchmark paragraph and section chunks against evaluation set. This is the concrete setup for choose chunk granularity based on retrieval questions, and measure answer precision and source coverage supplies a direct confirmation.
Incorrect Answers
A:Call approved embedding model and store VECTOR with matching dimension. This concerns comparing meaning across differently worded tickets. Here, benchmark paragraph and section chunks against evaluation set.
C:Configure external model with approved identity and secure URL. This concerns a SQL ai solution consumes an external model endpoint. Here, benchmark paragraph and section chunks against evaluation set.
D:Benchmark representative prompts on candidate models. This concerns comparing models on cost and latency. Here, benchmark paragraph and section chunks against evaluation set.
Question 799
A database contains sections and paragraphs as potential chunks. After implementing the design, which check most directly validates the expected behavior?
- Measure retrieval precision on title-only versus body questions
- Run job twice and confirm no duplicate chunks
- Compare outputs to source IDs and model version
- Measure answer precision and source coverage
Explanation
Correct Answer
D:Measure answer precision and source coverage. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that chunk size was selected without relevance testing.
Incorrect Answers
A:Measure retrieval precision on title-only versus body questions. This concerns vectors for both short titles and long bodies. Instead, measure answer precision and source coverage.
B:Run job twice and confirm no duplicate chunks. This concerns being replayable after a crash. Instead, measure answer precision and source coverage.
C:Compare outputs to source IDs and model version. This concerns being embedded efficiently. Instead, measure answer precision and source coverage.
Question 800
A database contains sections and paragraphs as potential chunks. A test instead finds: Broad chunks produce irrelevant neighboring content. Which underlying issue best explains this finding?
- Pipeline generates new random chunk IDs each run
- Chunk size was selected without relevance testing
- Vector rows lack stable provenance metadata
- Model endpoint permissions or identity binding are absent
Explanation
Correct Answer
B:Chunk size was selected without relevance testing. It accounts for the specific observed failure: broad chunks produce irrelevant neighboring content. The intended solution is choose chunk granularity based on retrieval questions with the required setup.
Incorrect Answers
A:Pipeline generates new random chunk IDs each run. This concerns rerun doubles search documents. Here, chunk size was selected without relevance testing.
C:Vector rows lack stable provenance metadata. This concerns search result cannot be cited back to document. Here, chunk size was selected without relevance testing.
D:Model endpoint permissions or identity binding are absent. This concerns SQL calls fail with unauthorized errors. Here, chunk size was selected without relevance testing.
Developing AI-Enabled Database Solutions (DP-800) Practice Tests