Optimise total return and you get a concentrated, high-leverage path. Optimise sharpe-ratio and you get something that hates volatility, including upside volatility. Optimise profit-factor and you can get a system with eleven trades. Optimise return divided by max-drawdown and you get something tuned to one historical drawdown that will not repeat.
Two habits help. Add a penalty for low trade counts, for example requiring at least 100 trades or scaling the score by the square root of the count. And optimise a blend, such as the median Sharpe across five data subsets minus a turnover penalty.
Write down what you would refuse to trade before you optimise. If a 60% annual return with a 45% drawdown is unacceptable, encode that as a constraint rather than discovering it afterwards.
Related: parameter-optimisation, sharpe-ratio