1. Package a model without executable deserialization
The model uses distance_km and load_ratio and predicts hours. The artifact includes schema_version, model_id, coefficients, training_rows and bounds. It is a transparent learning package, not an approved operational model.

2. Validate request values and response meaning
The 100 km/0.5 load input returns a numeric prediction. Negative distance, boolean distance and missing load are rejected. No request can select another model or change an entitlement. The response is an estimate for a learning exercise.

3. Separate local serving from website deployment
The default binding is 127.0.0.1, not a public interface. The course PHP runtime serves learning files only. Public serving needs an independent approved deployment; no cloud workload or model endpoint is created by this task.
