This formula obviously breaks in boundary conditions. For example, there are players I've never played with, but feel like my meta reliability at reading them is more than 0% (I can practice reading players I'm not playing with by reading through games and trying to guess their alignment). Thus, I suspect the formula is wrong.
It also implies that if two brand new newbies are playing their first two games with each other in parallel (say they both joined both games available in the Micro queue), then as soon as one of the games ends, those players will have a 100% reliable meta read on each other within the other game. Sure, that meta experience has to be helpful, but 100% reliable seems like a stretch. (In fact, I would argue that meta reliability is positively, not negatively, correlated with
G
.)
Given that some quick checks show that this formula can't possibly be correct in all cases, and there are some suspicious parts in it (the constant factor in the exponent is exactly -1/365? seriously?), I suspect that the formula is just made up rather than fit to any actual data about meta reads or any mathematical argument about how accurate they should be in the abstract.