A walkthrough of how random forests handle missing data and enable sample clustering. Covers two scenarios: missing data in the training set and missing data in new samples to classify. The iterative approach uses proximity matrices built by tracking which samples land in the same leaf nodes across all trees. Proximity values are refined over multiple rounds until missing value guesses converge. A bonus insight shows that proximity matrices can be converted to distance matrices, enabling heat maps and MDS plots regardless of data type (numeric, categorical, ranked). For classifying new samples with missing values, two copies of the sample are created with different label assumptions, missing values are imputed iteratively, and the copy most consistently classified correctly by the forest wins.