Out-of-sample testing evaluates a strategy on data that was kept separate from the data used to build or optimise it. It helps reveal whether a strategy may be overfitted.