For example, you might make a (successful) bet that works only in a low-volatility environment; if it works, your ex post Sharpe ratio will look ingenious.
But, if a given fund (Medallion in this case) is able to thrive/survive during the 1998 Russian financial crisis (which felled LTCM), the 2000-2004 tech bubble bursting, the 2007 quant quake, the 2008 financial crisis, and 2010 flash crash, it would seem to suggest that their statistical arbitrage strategy performs well in all manner of environments.
After their very hefty 5/40 fees, their worst year between 2001 and 2013 was a 21% gain [1].
I understand that there are all manner of epistemic and mathematical problems that prevent us ever from completely disentangling luck and skill, but a conceptual framework like "The Superinvestors of Graham and Doddsville" would seem to apply here. I.e., after a long period of consistent overperformance, the case for skill starts to look much more likely.