What should you validate before you compare a backtest against a real baseline?

The most useful ways to evaluate compare a backtest against a real baseline are the ones that show whether it still works after costs, timing, turnover, and portfolio context are included, not just whether one chart looks cleaner in hindsight.

That usually means asking whether compare a backtest against a real baseline improves the decision after friction, not just whether it makes one in-sample score look impressive.

The right metrics depend on what compare a backtest against a real baseline is supposed to control, but they should all connect back to whether the downstream trading rule becomes more robust, more scalable, or more honest about its limits.

Even a clean evaluation can hide fragility if the benchmark is weak, the assumptions are too generous, or the apparent improvement disappears once research workflow has to interact with the rest of the portfolio. That is why evaluation should end with a simpler counterfactual, not just a better chart.