Robustness is tested by deliberately trying to break the result. Shift the start date by three months. Shift every entry by one bar. Double the assumed costs. Drop the best 5% of trades. Run it on a related instrument it was not designed for.
A robust strategy gets worse under each of these and stays positive. A fragile one has one test that destroys it. Finding which test destroys it tells you what the strategy actually depended on, which is often something you did not intend.
Example: a futures system returning 14% a year that falls to minus 2% when you delete its ten best days was never a trend system; it was a handful of gap events, and its risk is entirely about whether those recur.
Related: parameter-sensitivity, monte-carlo-simulation, stress-test