Curve fitting
Tuning a strategy's rules until they fit a particular stretch of historical data unusually well, at the cost of working on new data.
Every rule, filter or parameter added to a strategy gives it another chance to fit noise in the data it was built on rather than a real, repeatable edge. With enough parameters, almost any past data can be fit closely.
A curve-fit backtest often shows an equity curve that is suspiciously smooth: a near-straight line with very small, evenly spaced dips is a pattern real trading, with its unpredictable market, rarely produces on its own.