Held-out results
Choose rules on an earlier training period, freeze them, and evaluate a later period excluded from that choice. Save both results, including failures.
Once you adjust the strategy in response to the held-out result, that period becomes part of development. A later untouched test is needed to evaluate the revision.
Parameter sensitivity
Compare nearby values using the same dates, sizing and costs. Record the number of alternatives tried, including abandoned variants.
A narrow peak is a reason to investigate. A broad plateau can still be selected by chance and does not certify future performance.
Cost and fill sensitivity
Repeat with plausible commission and slippage assumptions, then inspect individual orders. Keep the assumptions consistent across training and validation.
A cost change can expose a fragile result without proving overfitting. Bad data, unrealistic fills and changing market conditions are separate possible causes.
Trade concentration
Inspect how much of the result comes from a few trades, one direction or a short period. Review average wins and losses alongside win rate.
There is no universal trade-count, win-rate or performance-drop cutoff that establishes a reliable strategy.