Curve fitting
Tuning rules to the accidental features of a historical sample: the more variants are tried, the better the result on the past looks and the less it says about the future.
How to read the number
The giveaway is fragility: a small change of period, threshold or universe reverses the outcome. A durable pattern does not behave that way.
When the metric lies
The number of variants tried is almost never reported, and without it significance cannot be judged: with enough searching an outstanding result appears even on a random series.
Also known as: fitting rules to history