Risk of Over-Optimization

How do you know when you’ve tweaked a strategy so much that it only works on past data and fails in live markets?