Whenever you're ready, walk me through how you'd approach this.
How do you differentiate between true signal and noise in a high-dimensional feature space?
Treat this as a hands-on modeling task. Describe and implement a defensible workflow that covers preprocessing, regularization, feature selection, validation, null or permutation testing, and out-of-sample evaluation. Explain how you would detect leakage, correlated redundant features, unstable predictors, overfitting, and performance degradation under distribution shift. State which metrics and diagnostics you would use to decide whether a feature or model adds genuine predictive value.