Question 450
An API receives tokens issued by the right tenant but intended for a different service; it must reject them for this API. What is the best primary mechanism to use?
- Managed identity to database
- CREATE SECURITY POLICY
- Inline table-valued predicate function
- TLS termination
- Per-role API operation policy
- Validate JWT audience
Explanation
Correct Answer
F:Checking a token's audience prevents treating tokens minted for unrelated resources as authorization to this endpoint.
Incorrect Answers
A:A managed identity supports token-based database sign-in if the SQL platform and database principal are configured appropriately; the missing behavior is Validate JWT audience; this mechanism does something else.
B:CREATE SECURITY POLICY associates filter and block predicates with a table and controls whether the policy is enabled; that function is not interchangeable with Validate JWT audience for the requested result.
C:RLS uses predicate functions to express per-row access rules evaluated by security policies, often with SCHEMABINDING; for the stated conditions, the deciding feature is Validate JWT audience rather than this behavior.
D:TLS secures transport confidentiality and peer authentication, while application roles still govern allowed operations; it addresses a different step; the question calls for Validate JWT audience.
E:Role-level operation restrictions at the API boundary can separate read capabilities from write mutations without confusing authentication with authorization; this differs from the Validate JWT audience requirement described in the scenario.
Question 451
Many database readers can access source rows, but only one vetted service may generate new embeddings through a costly model. What should the developer implement?
- Restricted model EXECUTE rights
- Endpoint rate-limit handling
- Sensitive-data minimization
- Managed identity to Azure AI
Explanation
Correct Answer
A:Model-specific EXECUTE permissions can limit who invokes the external model, separately from SELECT on its source tables.
Incorrect Answers
B:Backoff, rate-limit-aware batching, and idempotent checkpoints handle transient throttling while protecting upstream model capacity; the alternative cannot substitute for Restricted model EXECUTE rights under the given constraints.
C:Minimizing model inputs to approved fields limits unnecessary disclosure and reduces retention and compliance risk; the condition in the stem is resolved by Restricted model EXECUTE rights, not by this feature alone.
D:Managed identity authentication can replace static secrets for supported endpoints when the workload identity receives the required service roles; the missing behavior is Restricted model EXECUTE rights; this mechanism does something else.
Question 452
Queries increasingly use a different vocabulary after a product launch, and the team wants to check whether vector retrieval quality degraded on new examples. Which option satisfies the desired behavior most directly?
- Precision at k
- Latency percentile
- Recall at k
- Embedding drift check
Explanation
Correct Answer
D:Periodic evaluation on recent representative queries can detect semantic quality drift and trigger reindexing or model comparison.
Incorrect Answers
A:Precision@k measures relevance within the top-k returned results rather than the total pool's coverage; for the stated conditions, the deciding feature is Embedding drift check rather than this behavior.
B:Latency percentiles expose tail-response behavior that mean timings can conceal in production; it addresses a different step; the question calls for Embedding drift check.
C:Recall@k measures the fraction of relevant items retrieved in a top-k result and reveals missed evidence; this differs from the Embedding drift check requirement described in the scenario.
Question 453
A regulated multi-tenant RAG system must show precisely which approved sources support an answer and must exclude sources belonging to other tenants before the model is called. Which TWO controls are essential? Choose TWO.
- Tenant-safe retrieval
- Source-document identifiers
- Freshness-aware indexing
- Evidence-only prompt instruction
- Retrieval recall benchmark
Explanation
Correct Answers
A:Authorization-aware retrieval filters evidence before model input, preventing cross-tenant disclosure through generation; used in combination with Source-document identifiers, it handles a separate requirement of this combined workflow.
B:Keeping source and chunk IDs through retrieval and prompt construction supports traceability and user-verifiable citations; used in combination with Tenant-safe retrieval, it handles a separate requirement of this combined workflow.
Incorrect Answers
C:Updating or invalidating affected chunk embeddings keeps retrieved context aligned with source changes and effective dates; that function is not interchangeable with Tenant-safe retrieval and Source-document identifiers for the requested result.
D:Directing generation to use only retrieved evidence and abstain when evidence is missing reduces unsupported claims, though validation remains necessary; for the stated conditions, the deciding feature is Tenant-safe retrieval and Source-document identifiers rather than this behavior.
E:Measuring retrieval recall on labeled questions isolates missing-evidence failures from generation mistakes; it addresses a different step; the question calls for Tenant-safe retrieval and Source-document identifiers.
Question 454
A team operates an existing managed Azure web application with deployment slots and built-in web hosting. It wants to deploy its Data API builder container within that existing hosting service rather than provision a Kubernetes cluster or a new container platform. Which hosting choice fits?
- Azure Monitor metrics
- Local DAB runtime
- Azure Container Apps
- Azure Kubernetes Service
- Azure App Service
Explanation
Correct Answer
E:App Service provides managed web hosting and deployment features for supported DAB deployment patterns.
Incorrect Answers
A:Azure Monitor captures supported resource metrics and evaluates metric-based alert rules against them; this differs from the Azure App Service requirement described in the scenario.
B:Running DAB locally supports iterative testing and debugging using its configuration and accessible development data source; the required capability here is Azure App Service, which this operation does not supply.
C:Azure Container Apps can host a DAB container with managed ingress and scaling suitable for event-driven or API workloads; the alternative cannot substitute for Azure App Service under the given constraints.
D:AKS hosts containerized DAB workloads with Kubernetes orchestration under customer-managed cluster configurations; the condition in the stem is resolved by Azure App Service, not by this feature alone.
Question 455
A lawyer searches contracts for exact specialized terms and phrases and wants linguistic token-based matching, not vector meaning. Which database mechanism should be chosen?
- Full-text search
- Hybrid keyword-plus-vector search
- Semantic vector search
- Structured relational filter
Explanation
Correct Answer
A:SQL full-text search indexes terms and supports linguistic predicates such as CONTAINS and FREETEXT rather than numerical vector proximity.
Incorrect Answers
B:Hybrid retrieval combines lexical and embedding evidence, capturing both exact-term and conceptual relevance; for the stated conditions, the deciding feature is Full-text search rather than this behavior.
C:Vector search compares embedding representations and can retrieve semantic neighbors despite different literal words; it addresses a different step; the question calls for Full-text search.
D:Ordinary indexed SQL predicates are the direct mechanism for precise structured filtering without embedding model inference; this differs from the Full-text search requirement described in the scenario.
Question 456
A single SELECT must map numeric risk scores into 'Low', 'Medium', and 'High' based on ordered conditions. Which database mechanism should be chosen?
- CASE expression
- NULLIF
- TRY_CONVERT
- COALESCE
Explanation
Correct Answer
A:CASE evaluates conditional branches and yields a scalar expression selected from matching WHEN conditions.
Incorrect Answers
B:NULLIF returns NULL when its two expressions are equal, often used with division to handle zero denominators; the condition in the stem is resolved by CASE expression, not by this feature alone.
C:TRY_CONVERT attempts type conversion and returns NULL for many invalid values where a regular CONVERT would fail; the missing behavior is CASE expression; this mechanism does something else.
D:COALESCE returns the first non-NULL expression from its ordered arguments; that function is not interchangeable with CASE expression for the requested result.
Question 457
A schedule query needs to compare each service window against the next scheduled start time for the same technician. Which option satisfies the desired behavior most directly?
- LAG()
- SUM() OVER(PARTITION BY ...)
- FIRST_VALUE()
- LEAD()
Explanation
Correct Answer
D:LEAD reads a following row's value in the specified window partition and order.
Incorrect Answers
A:LAG accesses a prior row's expression value within the ordered window without a self-join; it addresses a different step; the question calls for LEAD().
B:A windowed SUM computes totals across partitions while retaining individual input rows in the output; this differs from the LEAD() requirement described in the scenario.
C:FIRST_VALUE returns the expression from the first row of the applicable ordered window frame; the required capability here is LEAD(), which this operation does not supply.
Question 458
A retrieval team wants to know how many known relevant support answers appear within the top ten vector search results. Which SQL capability is the most direct fit?
- Latency percentile
- Recall at k
- Precision at k
- Embedding drift check
Explanation
Correct Answer
B:Recall@k measures the fraction of relevant items retrieved in a top-k result and reveals missed evidence.
Incorrect Answers
A:Latency percentiles expose tail-response behavior that mean timings can conceal in production; the missing behavior is Recall at k; this mechanism does something else.
C:Precision@k measures relevance within the top-k returned results rather than the total pool's coverage; for the stated conditions, the deciding feature is Recall at k rather than this behavior.
D:Periodic evaluation on recent representative queries can detect semantic quality drift and trigger reindexing or model comparison; it addresses a different step; the question calls for Recall at k.
Question 459
A SELECT on the multi-tenant payments table should silently return only rows associated with the current authenticated tenant. Which database mechanism should be chosen?
- GRANT SELECT on view
- Masked column definition
- RLS block predicate
- RLS filter predicate
Explanation
Correct Answer
D:An RLS filter predicate limits visible rows for eligible operations at the table's security-policy boundary.
Incorrect Answers
A:Object-level SELECT permissions can be granted on the view while base-table privileges remain constrained, with ownership chaining considered; the required capability here is RLS filter predicate, which this operation does not supply.
B:A masking function on the column changes the result shown to principals lacking UNMASK, leaving the underlying value unchanged; the alternative cannot substitute for RLS filter predicate under the given constraints.
C:An RLS block predicate can reject changes that would violate a row-access policy during specified DML operations; the condition in the stem is resolved by RLS filter predicate, not by this feature alone.
Question 460
A tax calculation stores exact two-decimal monetary amounts and should not accumulate approximate binary floating-point error. Which option satisfies the desired behavior most directly?
- VARCHAR(80)
- DATE type
- DECIMAL(12,2)
- CHAR(2)
Explanation
Correct Answer
C:DECIMAL provides exact base-ten precision and a declared scale appropriate to financial arithmetic.
Incorrect Answers
A:VARCHAR holds variable-length non-Unicode text subject to the column's collation/code page, with less storage than NVARCHAR for many such values; for the stated conditions, the deciding feature is DECIMAL(12,2) rather than this behavior.
B:DATE represents a date without a time-of-day portion, simplifying comparisons that should ignore clock time; it addresses a different step; the question calls for DECIMAL(12,2).
D:CHAR stores fixed-length non-Unicode character data, with padding semantics for shorter supplied values; the required capability here is DECIMAL(12,2), which this operation does not supply.