when something significantly outperforms your baseline, don't just celebrate it; investigate it 2. **Isolate variables before forming conclusions** — "this video did well" tells you nothing; "this video did well because of X" tells you everything 3. **Design real tests** — hypothesis → experimental