Elo-based rankings are wiedely used in 1to1 rankings for chess and other sports. They take into consideration if a player is expected to win or to loose and compare the expectation with the real outcome. If the expectation fits the outcome, some points are added or substracted, if the expectation and the outcome differ, some points are added or substracted in order to have real outcome meet the expectaion for the next match. This may become really dynamic, as if you win and it was not expected given your past results, you will gain many points.

Elo-based ratings are statistical estimations of the outcome, where a player that was faster than you has performed better than you. We will use this knowledge to adopt the estimation for the future.

As we dont have 1:1 races (not really often, luckily) the elo-approach has to be adapted for x players. Let's say you have a race with 8 players, so player 1 basically races against all other players (2-8), player 2 races against all other players (1, 3-8) and so on.

Player 1 is expected to win against players 2 and 3 but is also expected to loose against 4-8. So we build up a sum of weighted expectation-realoutcomes, called Delta for every player in a race. That's what you find on a per-race-basis.

Hope this is understandable and fair enough for you all! 

In case of questions - please ask in the appropriate thread on the forums.

Information on elo-calculations may be found on wikipedia.