What is training-serving skew, and what strategies prevent it in production ML systems? How do feature stores and shared transformation code help maintain consistency?

Sample answer preview

Training-serving skew occurs when the features used for training differ from those used during inference, causing models to perform worse in production than in development.