How to evaluate a freqtrade strategy before trusting it

Read the source before the equity curve

• Disabled or huge stop-loss (stoploss = -0.99): the strategy survives by never admitting a loss.
• Dozens of hand-tuned magic numbers: likely overfit to one historical period.
• Look-ahead bias: any use of future candles (shift(-1)) invalidates the whole backtest.
• Unmaintained code targeting an old freqtrade API: it may not even run.

Backtest hygiene

• Model fees AND slippage — both, always.
• Multiple fixed windows across different regimes, agreed before you run anything.
• Multiple pairs: a strategy that only works on one coin is a coincidence, not a system.

Metrics that matter

• Max drawdown before total profit: it decides whether you can psychologically keep running it.
• Profit factor above ~1.3 across windows beats a spectacular single window.
• Trade count: hundreds of trades give statistical meaning; a handful do not.
• High win rate + rare huge losses = the classic trap. Check the loss tail.

Or skip the grind

This checklist is exactly what Vetta automates: AI source-code audits, fixed multi-year windows with fees and slippage, and honest risk labels on 300+ strategies. Read a full free sample report to see the output.