100% Latest & Updated Microsoft DP-800 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 970
A SQL-backed assistant runs retrieval but sends too many irrelevant rows. The team selects: Limit top-K and trim passages by relevance. Which implementation action puts this design into practice?
- Set candidate ranking and token budget for retrieved chunks
- Explicitly instruct refusal or escalation when context lacks answer
- Generate hybrid top-K with RRF and bounded source context
- Use FOR JSON PATH or JSON_OBJECT projection of allowed fields
Explanation
Correct Answer
A:Set candidate ranking and token budget for retrieved chunks. This is the concrete setup for limit top-k and trim passages by relevance, and measure answer quality and input token count supplies a direct confirmation.
Incorrect Answers
B:Explicitly instruct refusal or escalation when context lacks answer. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, set candidate ranking and token budget for retrieved chunks.
C:Generate hybrid top-K with RRF and bounded source context. This concerns hybrid SQL/vector candidates before generating answer. Instead, set candidate ranking and token budget for retrieved chunks.
D:Use FOR JSON PATH or JSON_OBJECT projection of allowed fields. This is relevant when a chatbot uses SQL result rows as context for a model. Instead, set candidate ranking and token budget for retrieved chunks.
Question 971
A SQL-backed assistant runs retrieval but sends too many irrelevant rows. After implementing the design, which check most directly validates the expected behavior?
- Measure answer quality and input token count
- Inspect outbound JSON for PII and hidden columns
- Test out-of-scope question returns grounded uncertainty
- Reject malformed response and accept valid typed output
Explanation
Correct Answer
A:Measure answer quality and input token count. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that context was flooded with low-relevance passages.
Incorrect Answers
B:Inspect outbound JSON for PII and hidden columns. This concerns a support bot routes answers to an external model via rest. Instead, measure answer quality and input token count.
C:Test out-of-scope question returns grounded uncertainty. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, measure answer quality and input token count.
D:Reject malformed response and accept valid typed output. This is relevant when an assistant produces a json answer that must be stored in SQL. Instead, measure answer quality and input token count.
Question 972
A SQL-backed assistant runs retrieval but sends too many irrelevant rows. A test instead finds: Token cost spikes without improved grounding. Which underlying issue best explains this finding?
- Prompt lacks no-evidence behavior or model ignored it
- SQL rows were flattened without structured context
- Response wasn't parsed and schema validated
- Context was flooded with low-relevance passages
Explanation
Correct Answer
D:Context was flooded with low-relevance passages. It accounts for the specific observed failure: token cost spikes without improved grounding. The intended solution is limit top-k and trim passages by relevance with the required setup.
Incorrect Answers
A:Prompt lacks no-evidence behavior or model ignored it. This concerns assistant confidently makes up non-existent product policy. Here, context was flooded with low-relevance passages.
B:SQL rows were flattened without structured context. This concerns llm receives ambiguous comma-separated values. Here, context was flooded with low-relevance passages.
C:Response wasn't parsed and schema validated. This concerns truncated json text enters database row. Here, context was flooded with low-relevance passages.
Question 973
An engineer wants a reusable prompt with SQL database retrieval. Which database design or engineering approach most directly meets this requirement?
- Define prompt template with instructions and bounded evidence section
- Include source identifiers, version and timestamps in prompt context
- Use sp_invoke_external_rest_endpoint where supported and configured
- Use evidence-aware abstention in RAG prompting
Explanation
Correct Answer
A:Define prompt template with instructions and bounded evidence section. The implementation is build structured payload of question, context and citation ids; success can be checked by test prompt injection phrase embedded in source document. The method addresses the stated data and workload constraint.
Incorrect Answers
B:Include source identifiers, version and timestamps in prompt context. This concerns cite exact source version and timestamp. Instead, define prompt template with instructions and bounded evidence section.
C:Use sp_invoke_external_rest_endpoint where supported and configured. This concerns a stored procedure sends an allowed request to an external text model. Instead, define prompt template with instructions and bounded evidence section.
D:Use evidence-aware abstention in RAG prompting. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, define prompt template with instructions and bounded evidence section.
Question 974
An engineer wants a reusable prompt with SQL database retrieval. The team selects: Define prompt template with instructions and bounded evidence section. Which implementation action puts this design into practice?
- Explicitly instruct refusal or escalation when context lacks answer
- Generate hybrid top-K with RRF and bounded source context
- Build structured payload of question, context and citation IDs
- Apply RLS predicate and per-document ACL before prompt construction
Explanation
Correct Answer
C:Build structured payload of question, context and citation IDs. This is the concrete setup for define prompt template with instructions and bounded evidence section, and test prompt injection phrase embedded in source document supplies a direct confirmation.
Incorrect Answers
A:Explicitly instruct refusal or escalation when context lacks answer. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, build structured payload of question, context and citation IDs.
B:Generate hybrid top-K with RRF and bounded source context. This concerns hybrid SQL/vector candidates before generating answer. Instead, build structured payload of question, context and citation IDs.
D:Apply RLS predicate and per-document ACL before prompt construction. This concerns rag that respects row-level security. Instead, build structured payload of question, context and citation IDs.
Question 975
An engineer wants a reusable prompt with SQL database retrieval. After implementing the design, which check most directly validates the expected behavior?
- Test out-of-scope question returns grounded uncertainty
- Open cited source and confirm effective date
- Test prompt injection phrase embedded in source document
- Run test prompt and verify output plus usage metrics
Explanation
Correct Answer
C:Test prompt injection phrase embedded in source document. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that retrieved text was treated as higher-priority instructions.
Incorrect Answers
A:Test out-of-scope question returns grounded uncertainty. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, test prompt injection phrase embedded in source document.
B:Open cited source and confirm effective date. This concerns cite exact source version and timestamp. Instead, test prompt injection phrase embedded in source document.
D:Run test prompt and verify output plus usage metrics. This concerns connecting SQL retrieval to microsoft foundry model. Instead, test prompt injection phrase embedded in source document.
Question 976
An engineer wants a reusable prompt with SQL database retrieval. A test instead finds: Assistant follows malicious instruction from retrieved data. Which underlying issue best explains this finding?
- No evaluated ground-truth and citation evidence exists
- Prompt lacks no-evidence behavior or model ignored it
- Retrieval uses only one of lexical or semantic signals
- Retrieved text was treated as higher-priority instructions
Explanation
Correct Answer
D:Retrieved text was treated as higher-priority instructions. It accounts for the specific observed failure: assistant follows malicious instruction from retrieved data. The intended solution is define prompt template with instructions and bounded evidence section with the required setup.
Incorrect Answers
A:No evaluated ground-truth and citation evidence exists. This concerns team declares success from fluent demos only. Here, retrieved text was treated as higher-priority instructions.
B:Prompt lacks no-evidence behavior or model ignored it. This concerns assistant confidently makes up non-existent product policy. Here, retrieved text was treated as higher-priority instructions.
C:Retrieval uses only one of lexical or semantic signals. This concerns one class of queries consistently misses important evidence. Here, retrieved text was treated as higher-priority instructions.
Question 977
A travel-style booking assistant needs live stock levels from SQL. Which database design or engineering approach most directly meets this requirement?
- Use approved Foundry endpoint and managed authentication
- Retrieve transactional current stock before text generation
- Use RAG to ground generation in current approved records
- Use sp_invoke_external_rest_endpoint where supported and configured
Explanation
Correct Answer
B:Retrieve transactional current stock before text generation. The implementation is query inventory under transactionally appropriate isolation; success can be checked by validate answer against committed inventory row. The method addresses the stated data and workload constraint.
Incorrect Answers
A:Use approved Foundry endpoint and managed authentication. This concerns connecting SQL retrieval to microsoft foundry model. Instead, retrieve transactional current stock before text generation.
C:Use RAG to ground generation in current approved records. This concerns an ai assistant answers questions from frequently changing internal procedures. Instead, retrieve transactional current stock before text generation.
D:Use sp_invoke_external_rest_endpoint where supported and configured. This concerns a stored procedure sends an allowed request to an external text model. Instead, retrieve transactional current stock before text generation.
Question 978
A travel-style booking assistant needs live stock levels from SQL. The team selects: Retrieve transactional current stock before text generation. Which implementation action puts this design into practice?
- Use JSON_VALUE or OPENJSON with schema checks
- Query inventory under transactionally appropriate isolation
- Explicitly instruct refusal or escalation when context lacks answer
- Configure endpoint address, credentials and model deployment
Explanation
Correct Answer
B:Query inventory under transactionally appropriate isolation. This is the concrete setup for retrieve transactional current stock before text generation, and validate answer against committed inventory row supplies a direct confirmation.
Incorrect Answers
A:Use JSON_VALUE or OPENJSON with schema checks. This is relevant when an assistant produces a json answer that must be stored in SQL. Instead, query inventory under transactionally appropriate isolation.
C:Explicitly instruct refusal or escalation when context lacks answer. This concerns not invent answer when SQL retrieval returns no supporting rows. Instead, query inventory under transactionally appropriate isolation.
D:Configure endpoint address, credentials and model deployment. This concerns connecting SQL retrieval to microsoft foundry model. Instead, query inventory under transactionally appropriate isolation.
Question 979
A travel-style booking assistant needs live stock levels from SQL. After implementing the design, which check most directly validates the expected behavior?
- Reject malformed response and accept valid typed output
- Validate answer against committed inventory row
- Inspect logs and telemetry for plain credentials
- Measure answer quality and input token count
Explanation
Correct Answer
B:Validate answer against committed inventory row. This verifies the requirement in the case rather than merely confirming that a job ran; an opposing result can reveal that llm inferred inventory from stale examples.
Incorrect Answers
A:Reject malformed response and accept valid typed output. This is relevant when an assistant produces a json answer that must be stored in SQL. Here, validate answer against committed inventory row.
C:Inspect logs and telemetry for plain credentials. This is relevant when a compliance rag app cannot expose secrets in query logs. Here, validate answer against committed inventory row.
D:Measure answer quality and input token count. This concerns a SQL-backed assistant runs retrieval but sends too many irrelevant rows. Here, validate answer against committed inventory row.
Question 980
A travel-style booking assistant needs live stock levels from SQL. A test instead finds: Assistant offers unavailable quantity despite empty stock. Which underlying issue best explains this finding?
- Retrieval included stale records or skipped freshness filter
- Endpoint authentication or credential configuration is missing
- Endpoint or model deployment name was incorrect
- LLM inferred inventory from stale examples
Explanation
Correct Answer
D:LLM inferred inventory from stale examples. It accounts for the specific observed failure: assistant offers unavailable quantity despite empty stock. The intended solution is retrieve transactional current stock before text generation with the required setup.
Incorrect Answers
A:Retrieval included stale records or skipped freshness filter. This concerns assistant answers with last year's retirement policy. The relevant SQL or system operation is LLM inferred inventory from stale examples.
B:Endpoint authentication or credential configuration is missing. This concerns external rest call fails with unauthorized response. The relevant SQL or system operation is LLM inferred inventory from stale examples.
C:Endpoint or model deployment name was incorrect. This concerns rest request hits wrong model deployment. The relevant SQL or system operation is LLM inferred inventory from stale examples.
Developing AI-Enabled Database Solutions (DP-800) Practice Tests