A single change can be a genuine improvement. But repeatedly changing stop rules, entry conditions, time windows, or the asset universe after seeing the results eventually fits noise in the historical sample rather than market structure.
The more degrees of freedom you have, the easier it is to find a combination that looks good by chance.
Five entry rules, five exit rules, multiple horizons, and multiple instruments can generate a huge number of candidate combinations. If you keep only the one with the highest return, the odds rise that you selected noise rather than a real edge.
Better in-sample performance should increase skepticism, not reduce it.
The more often you revise a strategy while looking at the same data, the less that data behaves like an independent test. Freeze the rule, then check whether it survives on an unseen out-of-sample period.
Complex rules need a reason to exist
Every added condition should have a defensible link to market structure. If complexity keeps increasing only because it improves historical performance, compare the strategy against a simpler baseline.
The revision history is part of the research record
Recording each strategy version, the reason for the change, pre-change results, and the next validation window makes it easier to distinguish genuine improvement from after-the-fact tuning.