Meridian Line's platform team wants every request and response served by the port-operations copilot's Model Serving endpoint to be automatically captured for later quality analysis, without adding custom logging code inside the application itself. What Model Serving feature provides this?

Inference tables are the Model Serving feature built specifically for this requirement, once enabled on an endpoint they automatically log every request payload and the corresponding response to a Delta table, with no application-level logging code required, which is exactly what supports later quality analysis, monitoring, or building an evaluation set from real production traffic. Provisioned throughput is a capacity and performance feature, reserving dedicated throughput for consistent latency, it has no role in capturing request or response content for analysis. Traffic splitting controls what fraction of incoming requests get routed to which of two or more deployed model versions, which supports gradual rollout, but it does not log anything, it only decides where a request goes. Scale-to-zero is a cost-control behavior letting an idle endpoint release its provisioned capacity, which is unrelated to whether requests and responses get captured for analysis, an endpoint can scale to zero with inference tables enabled or disabled independently. Exam framing: inference tables are the specific mechanism for automatic, code-free request/response logging on a Model Serving endpoint, distinct from throughput, traffic-splitting, or scaling features.