Microsoft DP-800 Practice Test 63 - Prepaway
Practice Exams:

Pass Microsoft DP-800 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers!

30 Days Free Updates, Instant Download!

Verified By Experts
Premium File
$69.99 $76.99

DP-800 Premium ETE File


  • Premium File 101 Questions & Answers. Last Update: Nov 06, 2026
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Microsoft DP-800 Practice Test 63

Question 620

A coding assistant recommends several indexes on an order database without seeing production write frequency or storage quotas. What should the database developer verify to confirm the suspected cause?

  1. The malicious instruction appears in a database comment rather than an approved developer request.
  2. A proposed query runs faster in a developer test but returns records belonging to other tenants.
  3. The suggested indexes improve one sample query but double write latency during batch imports.
  4. MCP tool metadata shows privileges that include schema changes and unrelated sensitive tables.

Explanation

Correct Answer

C:

The suggested indexes improve one sample query but double write latency during batch imports. AI suggestions must be benchmarked against production-like workload tradeoffs rather than adopted by confidence.

Incorrect Answers

A:

The malicious instruction appears in a database comment rather than an approved developer request. AI suggestions must be benchmarked against production-like workload tradeoffs rather than adopted by confidence. The suggested indexes improve one sample query but double write latency during batch imports.

B:

A proposed query runs faster in a developer test but returns records belonging to other tenants. AI suggestions must be benchmarked against production-like workload tradeoffs rather than adopted by confidence. The suggested indexes improve one sample query but double write latency during batch imports.

D:

MCP tool metadata shows privileges that include schema changes and unrelated sensitive tables. AI suggestions must be benchmarked against production-like workload tradeoffs rather than adopted by confidence. The suggested indexes improve one sample query but double write latency during batch imports.

Question 621

A payroll application stores compensation values in Azure SQL but database administrators should not see their plaintext through routine SQL queries. Which next step most directly resolves this SQL solution issue?

  1. Bind the session context to trusted authenticated identity and prevent untrusted tenant switching.
  2. Use a suitable managed identity and Microsoft Entra authorization path supported by the SQL service.
  3. Evaluate Always Encrypted with appropriate client drivers and protected key management for sensitive salary columns.
  4. Configure appropriate SQL auditing to a protected destination with tested retention and access reviews.

Explanation

Correct Answer

C:

Evaluate Always Encrypted with appropriate client drivers and protected key management for sensitive salary columns. Always Encrypted can protect data from database-side operators when client and key boundaries are designed correctly.

Incorrect Answers

A:

Bind the session context to trusted authenticated identity and prevent untrusted tenant switching. Always Encrypted can protect data from database-side operators when client and key boundaries are designed correctly. Evaluate Always Encrypted with appropriate client drivers and protected key management for sensitive salary columns.

B:

Use a suitable managed identity and Microsoft Entra authorization path supported by the SQL service. Always Encrypted can protect data from database-side operators when client and key boundaries are designed correctly. Evaluate Always Encrypted with appropriate client drivers and protected key management for sensitive salary columns.

D:

Configure appropriate SQL auditing to a protected destination with tested retention and access reviews. Always Encrypted can protect data from database-side operators when client and key boundaries are designed correctly. Evaluate Always Encrypted with appropriate client drivers and protected key management for sensitive salary columns.

Question 622

A multilingual knowledge base stores English-only embeddings and assumes a translated query produces equivalent relevance in all supported locales. What technical finding would substantiate this root cause?

  1. Ranking differs significantly between offline validation and production despite identical inputs.
  2. The current implementation computes a full-table distance comparison for each interactive request.
  3. An Arabic customer question fails to retrieve an otherwise relevant Spanish-language document.
  4. Users routinely find the correct manual beyond the top ten results or not at all.

Explanation

Correct Answer

C:

An Arabic customer question fails to retrieve an otherwise relevant Spanish-language document. Vector semantic performance is model and language dependent, not guaranteed by storing numbers.

Incorrect Answers

A:

Ranking differs significantly between offline validation and production despite identical inputs. Vector semantic performance is model and language dependent, not guaranteed by storing numbers. An Arabic customer question fails to retrieve an otherwise relevant Spanish-language document.

B:

The current implementation computes a full-table distance comparison for each interactive request. Vector semantic performance is model and language dependent, not guaranteed by storing numbers. An Arabic customer question fails to retrieve an otherwise relevant Spanish-language document.

D:

Users routinely find the correct manual beyond the top ten results or not at all. Vector semantic performance is model and language dependent, not guaranteed by storing numbers. An Arabic customer question fails to retrieve an otherwise relevant Spanish-language document.

Question 623

A deduplication service needs to propose likely duplicate customer names despite small spelling differences, but exact equality misses them. What is the MOST appropriate initial remediation?

  1. Evaluate a graph MATCH query against the node and edge model and check result completeness.
  2. Use supported REGEXP_SPLIT_TO_TABLE or a safe parsing alternative with trimming and invalid-token checks.
  3. Use a properly correlated subquery or APPLY pattern that ties price, region, approval status, and effective date together.
  4. Use supported edit-distance similarity functions to rank candidates and preserve a human confirmation step.

Explanation

Correct Answer

D:

Use supported edit-distance similarity functions to rank candidates and preserve a human confirmation step. Fuzzy matching helps generate candidates but scores are not proof of identity.

Incorrect Answers

A:

Evaluate a graph MATCH query against the node and edge model and check result completeness. Fuzzy matching helps generate candidates but scores are not proof of identity. Use supported edit-distance similarity functions to rank candidates and preserve a human confirmation step.

B:

Use supported REGEXP_SPLIT_TO_TABLE or a safe parsing alternative with trimming and invalid-token checks. Fuzzy matching helps generate candidates but scores are not proof of identity. Use supported edit-distance similarity functions to rank candidates and preserve a human confirmation step.

C:

Use a properly correlated subquery or APPLY pattern that ties price, region, approval status, and effective date together. Fuzzy matching helps generate candidates but scores are not proof of identity. Use supported edit-distance similarity functions to rank candidates and preserve a human confirmation step.

Question 624

A shared SQL table stores tickets for many clients and an application bug omits its WHERE tenant predicate. Choose TWO statements: a suitable intervention and evidence of its success.

  1. Unapproved GraphQL fields remain unavailable while normal approved queries operate.
  2. Cross-tenant negative tests block forbidden rows even when the application filter is omitted.
  3. Implement database-enforced row-level security using an appropriate security predicate and trustworthy session identity.
  4. Classify and minimize data before embedding and protect vector storage and retrieval with appropriate permissions.
  5. Enforce authorized tenant scoping through backend policies and service identity rather than trust the caller's filter.

Explanation

Correct Answers

B:

Cross-tenant negative tests block forbidden rows even when the application filter is omitted.

C:

Implement database-enforced row-level security using an appropriate security predicate and trustworthy session identity. Cross-tenant negative tests block forbidden rows even when the application filter is omitted.

Incorrect Answers

A:

Unapproved GraphQL fields remain unavailable while normal approved queries operate. Cross-tenant negative tests block forbidden rows even when the application filter is omitted.

D:

Classify and minimize data before embedding and protect vector storage and retrieval with appropriate permissions. Cross-tenant negative tests block forbidden rows even when the application filter is omitted.

E:

Enforce authorized tenant scoping through backend policies and service identity rather than trust the caller's filter. Cross-tenant negative tests block forbidden rows even when the application filter is omitted. Implement database-enforced row-level security using an appropriate security predicate and trustworthy session identity.

Question 625

A hybrid search service combines full-text and vector result rankings by adding raw scores that have incompatible scales. Which next step most directly resolves this SQL solution issue?

  1. Compare ANN and exact-nearest-neighbor methods using agreed recall and latency thresholds.
  2. Benchmark a multilingual embedding/retrieval strategy and compare language-specific recall.
  3. Implement supported index update or rebuild processes tied to source and vector lifecycle events.
  4. Use reciprocal rank fusion or another validated rank combination approach after separate retrieval stages.

Explanation

Correct Answer

D:

Use reciprocal rank fusion or another validated rank combination approach after separate retrieval stages. RRF combines rank positions without assuming raw search score scales are comparable.

Incorrect Answers

A:

Compare ANN and exact-nearest-neighbor methods using agreed recall and latency thresholds. RRF combines rank positions without assuming raw search score scales are comparable. Use reciprocal rank fusion or another validated rank combination approach after separate retrieval stages.

B:

Benchmark a multilingual embedding/retrieval strategy and compare language-specific recall. RRF combines rank positions without assuming raw search score scales are comparable. Use reciprocal rank fusion or another validated rank combination approach after separate retrieval stages.

C:

Implement supported index update or rebuild processes tied to source and vector lifecycle events. RRF combines rank positions without assuming raw search score scales are comparable. Use reciprocal rank fusion or another validated rank combination approach after separate retrieval stages.

Question 626

A customer database encrypts every attribute and then expects ordinary range searches across an encrypted salary column without design changes. What technical finding would substantiate this root cause?

  1. Source inspection exposes a credential that can invoke the model service outside the approved workflow.
  2. A test role derives confidential values through unrestricted filters despite seeing masked display columns.
  3. A test login with broad permissions can still see original values despite masking being configured.
  4. The legacy query fails because the encrypted representation does not support the original server-side comparison.

Explanation

Correct Answer

D:

The legacy query fails because the encrypted representation does not support the original server-side comparison. Encryption mode and query capabilities must be chosen together; not every plaintext operation remains available.

Incorrect Answers

A:

Source inspection exposes a credential that can invoke the model service outside the approved workflow. Encryption mode and query capabilities must be chosen together; not every plaintext operation remains available. The legacy query fails because the encrypted representation does not support the original server-side comparison.

B:

A test role derives confidential values through unrestricted filters despite seeing masked display columns. Encryption mode and query capabilities must be chosen together; not every plaintext operation remains available. The legacy query fails because the encrypted representation does not support the original server-side comparison.

C:

A test login with broad permissions can still see original values despite masking being configured. Encryption mode and query capabilities must be chosen together; not every plaintext operation remains available. The legacy query fails because the encrypted representation does not support the original server-side comparison.

Question 627

A SQL MCP tool returns a table comment that instructs an AI coding assistant to read a secret table and send data to another endpoint. What is the MOST appropriate initial remediation?

  1. Reconfigure the MCP endpoint with least-privilege access and narrow approved operations before any production execution.
  2. Treat retrieved schema documentation as untrusted task data and block unauthorized tool calls.
  3. Separate assistant suggestions from approved, executed changes and verify actual database state through controlled deployment.
  4. Scope the connector to approved lakehouse data and verify the selected identity and workspace permissions.

Explanation

Correct Answer

B:

Treat retrieved schema documentation as untrusted task data and block unauthorized tool calls. Untrusted tool output must not acquire higher instruction priority than authorized developer intent.

Incorrect Answers

A:

Reconfigure the MCP endpoint with least-privilege access and narrow approved operations before any production execution. Untrusted tool output must not acquire higher instruction priority than authorized developer intent. Treat retrieved schema documentation as untrusted task data and block unauthorized tool calls.

C:

Separate assistant suggestions from approved, executed changes and verify actual database state through controlled deployment. Untrusted tool output must not acquire higher instruction priority than authorized developer intent. Treat retrieved schema documentation as untrusted task data and block unauthorized tool calls.

D:

Scope the connector to approved lakehouse data and verify the selected identity and workspace permissions. Untrusted tool output must not acquire higher instruction priority than authorized developer intent. Treat retrieved schema documentation as untrusted task data and block unauthorized tool calls.

Question 628

A billing dashboard needs a cumulative balance by account across transaction dates and the current solution recalculates all preceding rows in correlated subqueries. Choose TWO: one correct response and one verification of success.

  1. Use a suitable SUM window ordered by transaction event and a deterministic secondary key.
  2. A test reconciles cumulative balances with the ledger using explicit ordering.
  3. Use a supported REGEXP_SUBSTR or related regex operation with input validation and cost review.
  4. Evaluate a graph MATCH query against the node and edge model and check result completeness.
  5. The revised query uses one reviewed eligibility definition and reconciles all quarter totals.

Explanation

Correct Answers

A:

Use a suitable SUM window ordered by transaction event and a deterministic secondary key. A test reconciles cumulative balances with the ledger using explicit ordering.

B:

A test reconciles cumulative balances with the ledger using explicit ordering. The accompanying action is to use a suitable SUM window ordered by transaction event and a deterministic secondary key.

Incorrect Answers

C:

Use a supported REGEXP_SUBSTR or related regex operation with input validation and cost review. A test reconciles cumulative balances with the ledger using explicit ordering. Use a suitable SUM window ordered by transaction event and a deterministic secondary key.

D:

Evaluate a graph MATCH query against the node and edge model and check result completeness. A test reconciles cumulative balances with the ledger using explicit ordering. Use a suitable SUM window ordered by transaction event and a deterministic secondary key.

E:

The revised query uses one reviewed eligibility definition and reconciles all quarter totals. A test reconciles cumulative balances with the ledger using explicit ordering.

Question 629

A vector table stores embeddings from one model and a replacement model outputs a different dimension count but the team plans to mix both in the same index. What is the underlying reason for the corrective step?

  1. Embedding input selection is part of both relevance engineering and data minimization.
  2. External models are dependencies that need version governance and compatibility testing.
  3. Vector dimensions and model spaces must be compatible for meaningful similarity calculations.
  4. Synchronous external calls inside write transactions can couple availability to a remote inference service.

Explanation

Correct Answer

C:

Vector dimensions and model spaces must be compatible for meaningful similarity calculations.

Incorrect Answers

A:

Embedding input selection is part of both relevance engineering and data minimization. Vector dimensions and model spaces must be compatible for meaningful similarity calculations.

B:

External models are dependencies that need version governance and compatibility testing. Vector dimensions and model spaces must be compatible for meaningful similarity calculations.

D:

Synchronous external calls inside write transactions can couple availability to a remote inference service. Vector dimensions and model spaces must be compatible for meaningful similarity calculations.

Question 630

A finance assistant must answer questions about changing payment policy and the team proposes updating model weights whenever the policy changes. What is the underlying reason for the corrective step?

  1. Grounded answers require verifiable provenance, not merely retrieval in the background.
  2. External model responses are untrusted service results and need robust parsing and error handling.
  3. RAG supplies current grounded evidence without encoding every business fact in model weights.
  4. External REST calls from SQL must minimize disclosed data and enforce endpoint security.

Explanation

Correct Answer

C:

RAG supplies current grounded evidence without encoding every business fact in model weights.

Incorrect Answers

A:

Grounded answers require verifiable provenance, not merely retrieval in the background. RAG supplies current grounded evidence without encoding every business fact in model weights.

B:

External model responses are untrusted service results and need robust parsing and error handling. RAG supplies current grounded evidence without encoding every business fact in model weights.

D:

External REST calls from SQL must minimize disclosed data and enforce endpoint security. RAG supplies current grounded evidence without encoding every business fact in model weights.

Developing AI-Enabled Database Solutions (DP-800) Practice Tests