Start with an idea you can explain in a sentence — for example, ‘horses with a top-two speed figure who are drawn low in five-furlong handicaps’. If you cannot explain why the edge should exist, no amount of filtering will save it.
Keep the rule count low. Every extra filter shrinks the sample and increases the chance you are describing noise rather than an edge. Three or four rules over several thousand qualifiers beats ten rules over eighty.
Test across at least three seasons and check the profit is not concentrated in a handful of big-priced winners. Look at P/L at both SP and BSP — a system that only works at Betfair SP still works, but you need to know that before you stake it.
Then paper-trade it forward for a month. Systems that survive that step tend to survive the year, and the ones that do not cost you nothing.
this is a test comment