Why StandardScaler Broke My kNN Model in Production (And The Fix)

I spent 3 days debugging terrible model performance only to discover one preprocessing mistake. Here's the exact checklist I now use before every sklearn pipeline.