Punter76,

I don't know how big a database you have but if I'm testing or creating a system I don't use the whole database to test results on. Because I have a big one (not many places you can say that and not get censored :-) I tend to test say up to the end of 2002 data to create and then test my results on the last 3 years to see if it holds up on fresh data. Alternatively I create using say NSW races and test state by state to see if there is uniformity across the country (not quite so good theoretically as there may be differences in the way races operate, track condition, etc. state to state). If the data you created the rules with give much better results than the data you do the final test on it looks like you backfitted.

KV