If I train (Decision tree or Perceptron) with Spread = RollLong = RollShort = Commission = Slippage = 0 and then train with set(BINARY), LossPayout=0, WinPayout = 80, I get the same test results. In this case with the 0 spread and 0 commission version a "profitable" strategy with a 42% win rate, and with the binary version a shitshow because of the 42% win rate. I'm using some combination of distance from the mean, a stochastic oscillator, and the average true range, and I'd at expect it to at least be able to spit out a 54-55% win rate.

Does Zorro not take the binary setting into consideration? Is there a way to target highest win rates in training? or am I just not doing this right?