How does a system created by the SEC detect that person A is a (close) friend of person B, with accuracy? Family I can understand, if you share the same last name etc, but friends?
However, I'm not sure they do have that information. I would speculate that the number of people who make large option bets before earnings correctly on the same stock over multiple quarters should be zero and if it is very small, it is worth investigating. If they see that a subset of traders is 3 people with Korean names, that would immediately suggest that something is wrong...
But honestly, it’s much more likely that it was simply the timing of the trades and the earnings reports multiple times over a couple years raised some eyebrows.