The Importance of Out-of-Sample Testing

How does testing a strategy on data excluded from development help evaluate whether its results may generalise?