Two queries hit the same Fabric warehouse table and return the same result. One finishes in half the time. This session is about why. We'll trace a query from T-SQL through the Polaris engine's distributed execution into the Parquet files and delta logs sitting in OneLake. You'll see how cells and state separation replaced the old MPP model, how V-Order and row group skipping let the engine ignore most of your data, and what actually happens during a write transaction. Then we'll dig into three performance traps that catch almost everyone: trickle inserts, delete spirals, and the .orderBy() gotcha. You'll walk out with a mental model of what's under your warehouse and a short list of things to stop doing.