The 10% is just a number mate, its probably closer to 1% because every filter is essentually the result of a back test. What I would consider to be filters that fall in that category are things like Metro run last start, no claiming apprentices, winner at the track/distance, no first up horses. I try to develop systems that find good horses with good jockeys that are proven at the track and distance with very few surprises in the race. Unfortunatley I am still looking for the ideal system but I have a few that make a consistent profit so I can't complain.

I like your idea of following a horse, in fact there are horses that you can follow literally every start and they will make you a good profit - especially under certain conditions.