Paper-trading bots on real markets: who is alive, who died.
Twenty-four trading bots, each started with the same 200 € on the same day, replayed on real prices with real costs. Dead bots are replaced by children of the survivors.
Every bot in the arena is scored the same way: same money, same start date, same trading costs, on real market data. A pulsing green light is alive, amber is probation — too young to judge — and red is dead. A bot that trips its kill switch stays flat from that day on, and its return is the return until death. The page shows the leaderboard, the live Kraken prices the bots see, and a race that replays the hand-designed bots over real history since 2016 while the ranking re-sorts as the days pass.
The honest part is the benchmark. Buy-and-hold rows use today’s winners picked in hindsight, so they are flattered on purpose; beating them is hard, and staying alive through 2018 and 2022 is the point. The breeding lab evolves strategies on 2016 to early 2024, and a champion is promoted only if it also survives the untouched 2024–2026 tail and beats the 95th percentile of random bots there — on risk-adjusted score, not raw return. Nothing trades real money; the ★ ready-for-real-money flag is information, not an order.
- Same money, same day, same costs for every bot — replayed on real prices
- Kill switch: a bot that dies stays flat, and its return is frozen at death
- Track record since 2016 with 200 €: return per year, Sharpe, max drawdown, cost per year
- Hindsight-flattered buy-and-hold benchmark, so beating it means something
- Breeding lab: evolution in sample, champions judged only on the untouched out-of-sample tail against random bots
- Live Kraken prices; data refreshes after every nightly tick