Have you tried working backwards.From a winner to the neural??I mean adjusting the settings to make a known winner your first or second rated horse.Then using those settings for future predictions.Nothing is failsafe,, but it has some merit.