Win rate stays gated until a real backtest over a large settled sample confirms positive expectancy after payout asymmetry and fees.