Curve fitting (here) is the abuse of filters. In general it is cherry picking data to prove a point.

This typically shows up when somebody wants to show better results on a re-run and would likely end up cutting a time period or even a full month out.

Curve fitting of course can become an issue with firward testing as well.

I had the ECHO prints have a perfect scoreboard for the last few months, but the last one was a dud. You can never be certain that you have all the necessary filters already in place.

You have to evaluate what was different on this occassion. Then comes the need experience when picking the right way of excluding this prompt.

For instance, the width of the Bollinger Bands could often help, but not here.

The two choices are: eliminate the fractal where the candle 2-hours prior spiked outside the upper BB or eliminate when the candle 2-hours prior had a range of 50+ pips. I went with the latter as the second case would happen way less often.