How could one establish it in any other way besides going out and testing its testable predictions?

Using a notion as the basis for strategy is a different thing than testing the notion, even if you end up performing the same externally visible actions - in the latter case, you're paying a lot more effort in keeping track of just what you're doing differently and how it's working, and ideally arranging for replication and coordination with other people following the strategy. Moreover, testing can be done in constructed environments with carefully selected cohorts rather than in semi-random environments with self-selected cohorts, which is what you get if people begin applying it as strategy. The latter case is a lot harder to work with, statistically speaking.