Whenever you're ready, walk me through how you'd approach this.
How do you handle non-stationary time series data in a production environment? Explain how you would detect changing distributions, prevent temporal leakage, choose features and models, validate performance under regime changes, and decide when to retrain or roll back. Address both gradual drift and abrupt structural breaks, including monitoring, alerting, backtesting, and safeguards against training-serving skew.