Checkpoint
One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.
Prove that HOW you split decides what you see. The data is 1..100 stored in sorted order. First split it naively (first 50 vs last 50) and print both means (:.1f). Then seed random with 7, shuffle a COPY, split that in half, and print those means. Finally print whether shuffling shrank the between-halves gap. The naive split isn't 'sampling noise' — it's bias from splitting ordered data.