http://en.wikipedia.org/wiki/Elo_rating_system#Mathematical_details

I have been reading about this. It seems that the general principle is to adjust R_a and R_b proportionately to the distance between them. If there is a big distance, then there is a greater adjustment. If there is a small distance, then there is a smaller adjustment.

It seems to me, however, that there need to be two seperate cases: when they are getting farther apart (what you would expect- the outcome confirms the initial ratings of one over the other), and when they are getting closer together (an "upset").

When they are getting farther apart, ELO seems to fail. If A is much greater than B and A wins, it should only move up a little and B should only move down a little. However according to ELO, and I am leaving out some constants,

R'(a)= R(a) K(S(a)-E(a)=R(a) K(S(a)-1/(1 dist(a,b)))

so the adjustment should be K(S(a)-1/(1 dist(a,b))) or taking out K,

S(a)-1/(1 dist(a,b)). This means that as the distance increases, the amount that R(a) changes increases. But if A already was way bigger than B, and A wins, then because of the huge distance, A should get wayyy bigger. This seems counter intuitive. Can someone explain this to me?