How We Test Bots
Every template and every bot you build goes through the same backtester. Here is exactly what it does and what it can't tell you.
Fills and costs
- Signals use only closed bars; entries fill at the next bar's open (no look-ahead).
- If a bar touches both the stop and the target, the stop is assumed to fill first. A bar that opens beyond the stop fills at that worse price.
- Every fill pays the exchange taker fee, half the spread, and market impact sized by the coin's recent dollar volume.
- Funding is charged or paid on every funding print while a position is open.
- Position size compounds from realized equity; total margin is capped. With isolated margin, a move past the liquidation distance before the stop counts as a liquidation.
- Delisted coins stay in the test, so the results aren't flattered by only using coins that survived.
- Equity is marked to market daily, so drawdowns include open-trade swings.
Checks before a "Passed out-of-sample" badge
- The last 30% of history is held out while a template is designed, then checked once.
- Out-of-sample: profitable, at least 30 trades, profit factor at least 1.1.
- Deflated Sharpe ratio at least 0.95, counting every variant tried while designing the bot.
- Luck test: the bot must beat at least 95% of random-entry versions with the same coins, sides and holding times.
- Max drawdown no worse than -60%. We also show walk-forward periods, bull/bear/chop results and simulated ruin odds.
What a backtest can't tell you
A backtest is hypothetical. It can't fully model fills during crashes, exchange outages, rule changes, or how you'll behave during a loss. Results that pass every check can still lose money live, and live results are usually worse. Read the disclaimers · How good is 5% a month?