I built a futures trading system with an overfitting lie-detector attached, then threw every popular trading idea at it — 1,900 real NASDAQ-futures sessions, honest train/test splits, no mercy. Eight chapters and a finale: what survived, what died, and why your backtest is probably lying to you.
The motivation, the rules I set before starting, and the process honestly told — including the adversarial review of my own code and the 90% kill rate. Start here.
PBO, deflated Sharpe, and combinatorial cross-validation explained for prop traders.
The volatility filter that passed every gate, and why it works.
Retest entries, volume confirmation, trend filters, tight stops — tested on 1,900 real sessions.
I swept every profit target. Higher win rates made less money. Here's the math.
Sizing can't help you pass an eval. It can transform a funded account. Monte Carlo proof.
Parameter stability heatmaps, regime-switching, and why your own research contaminates your backtest.
I pointed the machine at gold and crude: a data trap, an audit that shrank my own numbers, and why the index keeps the crown.
$12 of 1-minute data killed my best backtest ever, acquitted the champion at 92% confidence, and closed the research program.
76% win rate, beautiful curve, $0 commission column, and 59% of trades filled inside a single candle. Anatomy of a strategy that sells better than it trades.
Every return statistic for the strategy that passed my tests — distributions not averages — and the prop-firm ladder math that ruins the story.
Do big days announce themselves? 5,000+ sessions across index, gold and crude say yes — and that the coil-then-explode intuition is exactly backwards.
The whole program in six honest charts — equity, drawdowns, the win-rate trap visualized, the parameter-stability heatmap, and the out-of-sample proof.