How would you handle missing values?

An answer that works

First find out whether missingness is informative, and if so record it as a feature before imputing. Then drop rows missing something essential, and impute the rest with a median or a model, fitted inside the training fold to avoid leakage.