So you only use logical filters that aren't backfitted. Examples you give are - not first up - "d" - "t" - Metro run last start. For starters, I don't for the life of me know how you can suggest "d" and "t" are not backfitted. They are part of historical data (like a lot of other filters) which we use to predict future performance .....

Your filters, you suggest represent about 1%, because as you say every other filter is a result of a back test. So to find your filters we have to look for similarities to "d" and "t" etc. Days last start ??, nah can't be as some horses back up, some need a couple of weeks off. It's too individualistic. Race Prizemoney - ??, Weight ??, Barrier - that's one of them for sure. So we're looking for at least another 2 maybe 3 filters to bring a LOT of 20% into a POT of 20% (it's gotta be that much), and filters that aren't backfitted.

The difference between filters that are backfitted, and mattio's filters that aren't backfitted is - mattio's perception of what backfitting really is. I'm actually on his side in this and having thought about what he's saying, it's exactly the same path as I'm trying to go down. I do have a few good systems, and I believe they're logical (yup they contain filters that are backfitted, but I don't change them), but I also see the sense in what mattio's trying to convey here.