A. Incorrect.
GKE adds cluster management and baseline cost for a simple static site.

B. Incorrect.
Compute Engine does not scale to zero and requires VM management.

C. Correct.
Cloud Run is serverless, request-driven, autoscaling, and can scale to zero when not serving requests.

D. Incorrect.
Both GKE and Compute Engine add unnecessary management and ongoing cost for this scenario.

A. Incorrect.
This does not immediately change existing objects in the current bucket.

B. Incorrect.
A migration adds unnecessary steps and does not directly update the existing bucket.

C. Incorrect.
Lifecycle rules can help future transitions over time, but changing the default storage class is the direct control for future uploads.

D. Correct.
Rewriting changes the storage class of existing objects, and the bucket default controls new objects.

A. Incorrect.
A stateless MIG does not preserve the database disk state across recreation.

B. Incorrect.
A stateful MIG helps recovery, but Hyperdisk Extreme is the higher-performance disk option.

C. Incorrect.
This misses both the highest-performance disk and state preservation.

D. Correct.
Hyperdisk Extreme addresses high disk performance, and a stateful MIG preserves attached disk state during recovery.

A. Correct.
gcloud config set compute/zone europe-west1-d sets the default zone for CLI commands.

B. Incorrect.
The console setting does not reliably configure your local gcloud CLI default.

C. Incorrect.
Manually creating a file is not the supported configuration method.

D. Incorrect.
Project metadata does not set the gcloud CLI default zone.

A. Incorrect.
A Service exposes Pods; it does not define the Pods that run the container image.

B. Incorrect.
GKE pulls container images from a container registry, not generic Cloud Storage object paths.

C. Incorrect.
The registry is correct, but a Service alone does not run the workload.

D. Correct.
Store the image in a container registry and create a Deployment to manage the running Pods.

A. Correct.
Node auto-provisioning can add suitable GPU capacity when needed and remove unused capacity, minimizing idle GPU cost.

B. Incorrect.
VerticalPodAutoscaler adjusts requests; it does not provision GPU nodes by itself.

C. Incorrect.
Preemptible or Spot VMs are risky for long-running non-restartable jobs.

D. Incorrect.
A minimum size of 1 keeps an expensive GPU node running even when jobs are infrequent.

A. Incorrect.
Billing export helps analysis but does not change who is charged.

B. Incorrect.
Support does not convert your personal card billing into the company's invoice workflow.

C. Incorrect.
Granting access to your personal billing account keeps charges in the wrong account.

D. Correct.
Linking projects to the company's billing account moves future charges to the company invoice.

A. Incorrect.
Organization-level Editor is extremely broad and can affect many projects.

B. Incorrect.
Basic Editor can gain permissions as services evolve and is broader than needed.

C. Correct.
A custom role scoped to the production project limits access to the exact permissions required.

D. Incorrect.
Organization-level scope grants the custom role more broadly than necessary.

A. Incorrect.
Cloud CDN improves delivery performance but does not correct object MIME metadata.

B. Incorrect.
Public access changes permissions, not browser handling.

C. Correct.
The Content-Type metadata tells browsers the object is a PDF so it can be rendered inline when supported.

D. Incorrect.
Labels are not HTTP object metadata and do not control browser behavior.

A. Incorrect.
Custom deletion automation is risky and not the recommended first control.

B. Incorrect.
Scanning resources manually does not enforce future IAM domain restrictions.

C. Incorrect.
Organization Policy prevents future violations but does not automatically clean existing IAM bindings.

D. Correct.
Apply the domain restriction for future enforcement and remediate existing mismatched access separately.