A. Incorrect.
Azure DNS resolves names and does not train models.

B. Correct.
Automated ML can explore algorithms and featurization for supported ML tasks.

C. Incorrect.
Firewall DNAT publishes inbound network access and does not select algorithms.

D. Incorrect.
Blob archive tier stores rarely accessed data and does not build models.

A. Incorrect.
A NAT gateway controls outbound networking and does not explain predictions.

B. Incorrect.
DNS CNAME records map names and do not interpret models.

C. Correct.
Interpretability tools explain feature importance and prediction drivers.

D. Incorrect.
Storage leases coordinate blob access and do not explain ML behavior.

A. Incorrect.
A spreadsheet is not a production inference endpoint.

B. Incorrect.
Private DNS can support name resolution but does not host predictions.

C. Incorrect.
Recovery Services vaults store backup data and do not serve ML models.

D. Correct.
Managed online endpoints provide HTTPS real-time inferencing for Azure ML models.

A. Correct.
Batch endpoints are designed for asynchronous, large-scale batch inferencing.

B. Incorrect.
DNS lookups do not score ML records.

C. Incorrect.
Bastion provides VM access and does not run batch scoring.

D. Incorrect.
WAF rules filter web requests and do not perform inference.

A. Incorrect.
SMB is a file sharing protocol and not an experiment tracking tool.

B. Incorrect.
Traffic Manager routes DNS traffic and does not track ML runs.

C. Correct.
MLflow tracking is integrated with Azure ML for run metadata, artifacts, and models.

D. Incorrect.
Policy exemptions bypass governance rules and do not store experiment metrics.

A. Incorrect.
A DNS label names an endpoint and does not define packages.

B. Incorrect.
An NSG rule controls network traffic, not Python dependencies.

C. Incorrect.
A CDN origin group routes content delivery traffic.

D. Correct.
Azure ML environments define software dependencies for reproducible training and inference jobs.

A. Incorrect.
DNS zone transfers copy DNS data and are unrelated to model tuning.

B. Correct.
Hyperparameter sweeps evaluate multiple parameter combinations against an objective metric.

C. Incorrect.
Blob soft delete protects deleted blobs and does not tune models.

D. Incorrect.
Key Vault purge deletes vault objects and is not an ML training strategy.

A. Correct.
F1 score balances precision and recall and is often useful for imbalanced classification.

B. Incorrect.
DNS TTL is not an ML evaluation metric.

C. Incorrect.
Public IP count does not measure model quality.

D. Incorrect.
Storage replication protects data and does not evaluate classifiers.

A. Incorrect.
Private DNS link count is not a model input distribution metric.

B. Incorrect.
Storage SKU does not detect ML drift.

C. Correct.
Data drift monitoring compares production input distributions with baseline data.

D. Incorrect.
Gateway subnet size is networking design information.

A. Incorrect.
MX records route email and do not store ML models.

B. Incorrect.
NSGs and route tables control network traffic.

C. Incorrect.
Anonymous folders are insecure and lack ML registry metadata.

D. Correct.
The Azure ML model registry tracks model assets and versions for deployment.