The distribution of incoming prediction features differs substantially from training data. Which condition is being observed?

1. Proof that the API endpoint is unreachable
2. Data drift
3. A code-formatting defect by definition
4. Guaranteed model improvement