Microsoft DP-800 Questions & Answers

JSON_VALUE extracts a scalar path, while OPENJSON can shred collections.

Using MATCH with explicit edge direction and aliases for the required path would not correct an API expects a single relational row converted into a JSON object with named fields. Use JSON_OBJECT with explicit key/value expressions where supported.

NULL in a NOT IN subquery can yield UNKNOWN and unexpectedly filter all rows.

A test environment with modified index triggers a drift report. However, that result would not establish the required outcome: drift detection flags the unexpected column change before publish.

After publish, the target contains the intended index and tested workload improves. However, that result would not establish the required outcome: drift detection flags the unexpected column change before publish.

DACPAC inspection contains the new approved procedure. However, that result would not establish the required outcome: drift detection flags the unexpected column change before publish.

Path ownership helps route high-impact DDL for accountable review. The actual technical constraint is that optional checks fail to enforce controls during urgent releases.

Schema reconciliation can propose destructive changes that threaten existing records. The actual technical constraint is that optional checks fail to enforce controls during urgent releases.

DDL diffs have operational consequences beyond text formatting. The actual technical constraint is that optional checks fail to enforce controls during urgent releases.

Maintain a controlled model catalog and audit changes. The current task instead requires this action: log tool invocations and routinely remove unused tools.

Use pre-send secret scanning and developer training. The current task instead requires this action: log tool invocations and routinely remove unused tools.

Do not grant production DDL permissions directly to AI tools.

Changing vector dimensionality does not directly diagnose the ANN candidate-search settings and may invalidate the current index.

Relaxing authorization risks exposing restricted content and does not provide an acceptable solution to recall regression.

An even narrower search can improve speed while worsening the very missing-neighbor defect being investigated.

A test completion appears under an authorized developer identity. However, that result would not establish the required outcome: an approved SQL workspace exposes supported Copilot features for a licensed tester.

A security review finds no production secrets in prompt logs or history. However, that result would not establish the required outcome: an approved SQL workspace exposes supported Copilot features for a licensed tester.

Test prompts contain representative synthetic values, not real identifiers. However, that result would not establish the required outcome: an approved SQL workspace exposes supported Copilot features for a licensed tester.

Vector dimensionality is fixed by the column and must match the query vector. The actual technical constraint is that external model quotas and transient errors require controlled retry behavior.

Embedding the wrong language source can degrade multilingual retrieval quality. The actual technical constraint is that external model quotas and transient errors require controlled retry behavior.

Modeling invocation can export inputs and incur charges, so object privileges matter concerns a different mechanism; the relevant technical point is that external model quotas and transient errors require controlled retry behavior.

Retest constraints after bulk imports and schema changes. The current task instead requires this action: review cascading deletes for unintended financial data loss.

Document a schema contract and version changes to payload format.

Monitor exhaustion and limit allocation rights. The current task instead requires this action: review cascading deletes for unintended financial data loss.

A split-to-table routine returns relational tokens rather than one long string.

Jaro-Winkler-style measures can treat transpositions and prefixes differently from raw edits.

String resemblance cannot guarantee two records represent the same person.

A throttled test recovers without duplicate vector records or budget spikes. However, that result would not establish the required outcome: all new embeddings insert without dimension errors and search quality is rechecked.

A change-feed pilot keeps vectors synchronized using the selected supported method. However, that result would not establish the required outcome: all new embeddings insert without dimension errors and search quality is rechecked.

A staging inference call succeeds with the new approved endpoint. However, that result would not establish the required outcome: all new embeddings insert without dimension errors and search quality is rechecked.

External model quotas and transient errors require controlled retry behavior. The actual technical constraint is that embeddings are computed numeric representations and are not created merely by defining a VECTOR column.

Unconstrained prose is fragile to parse and can omit required keys. The actual technical constraint is that embeddings are computed numeric representations and are not created merely by defining a VECTOR column.

Text-only models cannot directly interpret image content without a supporting vision pipeline. The actual technical constraint is that embeddings are computed numeric representations and are not created merely by defining a VECTOR column.