A regulated lender needs a durable record of every scoring call's input, output and deployed model version. Which endpoint capability provides this?
Show answer and explanation
Answer: D
Vertex AI endpoints offer request-response logging, which records prediction inputs and outputs along with model metadata and can export them to BigQuery for audit and analysis. This creates the durable, queryable history a lender needs to explain individual decisions. Disabling logging removes the record entirely, distributed traces capture timing rather than payloads, and DNS resolver logs show name lookups rather than scoring activity.
Why the other options are wrong
- A. Turning off logging saves a little overhead but destroys the audit trail the lender is required to keep for each decision.
- B. Tracing spans describe how long each step took across services and do not preserve the actual payload sent to or returned by the model.
- C. DNS resolver logs record hostname lookups, so they show that a connection was attempted but not what was scored or which model answered.