How Missing Values Distort PCA Embeddings (And The Fix)

My PCA visualization showed perfect clusters until I realized 20% of my data was missing. Here's how missing values silently break dimensionality reduction.