Trading bot performance can look impressive on a dashboard, but a high return number by itself tells almost nothing about the quality of the strategy.
The first metric I would check is maximum drawdown. A bot that earns 40% but experiences a 35% drawdown is very different from one that earns the same return with a much more controlled risk profile. For me, drawdown says more about how difficult the strategy would be to actually hold through than headline profit.
Win rate is useful, but it can also be misleading. A bot can win 80% of its trades and still lose money if the losing trades are much larger than the winners. That is why I prefer looking at the relationship between average profit and average loss rather than focusing on win rate alone.
Profit factor is another useful metric. It compares gross profit with gross loss and can give a clearer picture of whether the strategy has a real edge. A bot with a strong profit factor over a large sample is more interesting than one showing a few profitable trades.
Trade count matters too. A strategy tested on twenty trades is not nearly as convincing as one that has gone through hundreds of different market conditions. The larger the sample, the easier it is to judge whether performance may be repeatable rather than the result of luck.
I also want to see performance across different market regimes. A bot that worked only during a strong Bitcoin bull market may simply be benefiting from the underlying trend. It is much more useful to know how the strategy performed during sideways markets, corrections and periods of high volatility.
Fees and slippage need to be included in every result. Some backtests look profitable only because they assume perfect execution. In real markets, trading fees, spreads and slippage can destroy the edge of a high-frequency strategy.
Another metric I would watch is risk-adjusted return. Sharpe ratio, Sortino ratio or similar measures can help compare strategies that generate different returns with different levels of volatility. They are not perfect, but they provide more context than raw profit.
Live performance is ultimately more valuable than a backtest. Historical simulations are useful, but I would trust a bot much more if it has a transparent real-money track record over a meaningful period.
What metrics do you check before trusting a crypto trading bot?
Do you focus most on maximum drawdown, profit factor, win rate, risk-adjusted return, trade count or live performance?
And what is the minimum track record you would want to see before letting a bot trade real capital?