What is curve fitting (overfitting) in trading?

Curve fitting is tuning a strategy until it fits the historical data's random noise instead of a repeatable pattern. The backtest looks spectacular because the parameters were chosen — by you or an optimizer — precisely because they scored well on that exact data. The noise doesn't repeat; the performance doesn't either.

It is the default failure mode of strategy development, not an exotic one. When we grid-tested 5,720 configurations across 10 strategy families and 20 coins, the median in-sample Sharpe of attractive configurations roughly halved on held-out data, and one in six flipped to outright losses.

Detection is procedural: split the data chronologically, tune only on the first part, and judge only on the last. If in-sample and out-of-sample tell different stories, you fit the noise.