Basically what mattio is saying is if say one of your filters is a horse must have a winning % above 23.37% then you've clearly backfitted to get the 'best' result. However if your filter is it has to be above 20% or 25% then thats not so bad as your using general numbers and not backfitting to a precise number.