I Made 1,000,000 Trading Bots Fight to the Death

I created a million trading bots and let them trade six years of real futures prices. Every month the worst half died and the best 10% had children. 37 million bots later, I let the survivors trade 2026, a year none of them had seen. The champion lost money. The population didn't. Here's why both are true. Run it yourself: support my work and get access to all code and infrastructure and 50+ hours of educational video material (this analysis included): the evolution lab that breeds a million bots on your own markets, with the random twin test built in. 👉 https://lab.wisequant.app What I did - Bots: one of 10 futures (Nasdaq, S&P 500, Dow, Russell, 10y and 30y US bonds, oil, gold, euro, yen) + up to 3 of 136 conditions + long or short + a fixed holding time (15 min to 1 day) - Equal risk per trade: sized so a normal day moves it by about $1,000 (a typical oil trade ≈ $46,000, a 10y bond trade ≈ $290,000), commission plus 1 tick of slippage in and out - Real 15-minute prices, January 2020 to September 2026 - Walk-forward: bots are only judged on months that are already over; 2026 was locked until the end - Random twin: same market, same direction, same hours in the market, random times, no fees Results - Random bots: −$32,227 a year each, −$41 before costs; 1 in 8 made a profit; 8,870 went bankrupt in January 2020 from fees alone - Run 1 (most money survives): 90% short oil after Covid, then 90% long oil, then the Nasdaq, then gold; no timing skill ($579 vs $608 a month for the twin) - Run 2 (beat your twin survives): $239 vs −$128 a month for the twin in 2021–2025, +$368 a month of timing (luck about 1 in 50) - 2026: Nasdaq +22%; population +$5,726 vs twin +$93 (ahead in 8 of 9 months); top 100 +$13,302 vs +$1,573 - Repeated 4 more times with 10,000 bots: the timing population beat its twin in 2026 every time - Champion (born Dec 2025, 2025 backtest +$38,988): −$3,335 in 2026; its twin +$2,093