Out-of-sample

Testing a strategy on out-of-sample data — a period the rules were never tuned against — is the main defense against overfitting: a rule set that performs well on the data it was fit to and much worse on data it's never seen was likely fit to noise rather than a real edge. Walk-forward testing is one structured way to keep generating fresh out-of-sample periods rather than testing on the same held-out slice repeatedly.