I’ve often posted about skill ranking and matching. One important complaint I had about H4’s skill/matching turns out to be entirely incorrect. So I would like to fix that here.
First, TrueSkill. TrueSkill is composed of three numbers for each player:
…m = average rank estimate
…s = rank uncertainty
…r = displayed rank, calculated as r = m - k*s (where k is usually 3)
The difference between m and r is that m represents the system’s best estimate of your skill. On the other hand, r represents the number that the system is 99% sure that your actual skill is higher than. In other words, if you have a CSR of 25, the system is 99% sure your actual skill is at least 25.
My incorrect claim: H4 uses r for matching. This starts everyone in a list at 1 for matching purposes. That mashes all of the high and low skill players together, which takes a long time to sort out. It would be better - ranks would sort themselves out more quickly - if m were used.
What actually happens: Matching in H4 already uses m.
In other words, TrueSkill does use the appropriate skill estimate when doing matching. That means the answer to better matching has nothing to do with using the median estimate . . . since that is what is already [and correctly] done.
Apologies for the mistake. I may write more later when I have more time to give a better explanation of the issues with H4.
P.S. The other aspects of those skill / matching posts seem to be accurate - at least based on my current understanding - but I was completely wrong on this particular (and rather important) point.
