* Controlled risk + low drawdowns - high risk systems may look amazing on paper, but nobody can stomach as much drawdown as they think they can.
* Fully end-to-end backtested + rules-based. Statistical edge = good. Random idea = bad.
* No black boxes/ML - this is a hot take but I think most ML approaches will just overfit faster. When the biggest issue in backtesting is over fitting, ML just makes the over fitting more efficient and easier to justify
* Good Sharpe ratio - the saying is "you can only eat risk-adjusted return"
* Survives parameter robustness stress tests. If it only "works" with one value it's probably overfit
* Every rule has some sort of justification besides the backtest. It needs to make sense intuitively
* Based on actual academic research, such as different risk premia extraction. Short volatility strategies are a good example of this