4× the backtest profit — and no more trustworthy.Bitcoin / TetherUSBINANCE:BTCUSDTalgolegoI took a well-known free strategy for this market — Pivot Point SuperTrend by LonesomeTheBlue — and deliberately over-fit it, to show how little a big backtest number proves on its own. I nudged a few inputs and switched it to long-only. On ~9 years of 1-hour BTCUSDT data, published defaults → my tuned version: • Net profit: $19,674 → $83,077 (4.2×) • Profit factor: 1.03 → 1.44 • Win rate: 36% → 44% • Max drawdown: 122% → 31% Looks like a decisive upgrade. It isn't. Stress-test the two versions and the things that decide whether an edge is real barely moved: 1. Walk-forward (out-of-sample): efficiency 0.79 → 0.80. I made the backtest 4× more profitable and out-of-sample generalization didn't budge — the fingerprint of fitting to this exact history. 2. Trade concentration: remove the top 11 winning trades (of 207) and the "improved" version loses ~$60,800. Fifty-three percent of the gross profit comes from 11 trades. A durable edge is spread across hundreds. 3. Risk-rule survival (prop-firm-style drawdown limits): 0% pass rate — same as before. The equity path still breaks the rules that protect real capital. And the quiet one: long-only through a multi-year BTC uptrend. The best years are 2020 and 2024 — the obvious ones. A chunk of that 4× isn't edge, it's beta. Takeaway: a bigger backtest number is easy to manufacture; durability is not. Before you trust a curve, ask what happens out-of-sample, whether the profit survives removing a few trades, and whether it holds under real drawdown limits. Educational only. Not financial advice.