Alright, i give a very short description of the algorithm:

Its a more "conventional" approach this time. Testing suggested that the actual problem with "noob bashing" is, that some of the new players are drastically overrated when they join the ladder. So, the idea is to get them to their approximate rating asap, without impacting the "established" players a lot.

The core algorithm is still the same:

K=32 //zone compatible k
S=250 //zone compatible scaling

p = 1 / (1+10^(Rate_loser-Rate_winner)/S) //expected winning-chance for winner
K_multiplier = 1.0 //see later

RatingChange_winner = K_*K_multiplier*(1-p)
RatingChange_loser = -K*K_multiplier*(1-p)

About the K_multiplier. This is determined for every player, and this means that the points won by one side are not always equal to the points lost on the other.

For every player, it gets determined by the following algo:

1) For high rated players, we use lower K-factors.
myrating>2100 -> K_multiplier=0.5
myrating>2350 -> K_multiplier=0.25

2) ELSE, If this player is new, use higher K-factors.
mynumbergames<12 -> K_multiplier=4.0
mynumbergames<24 -> K_multiplier=1.5

3) ELSE, If this player already has an "established" rating, and opponent is new, use lower K-factor for the established player.
opponentgames<24 -> K_multiplier = 0.5;

4) If none of the above matches, use
default -> K_multiplier=1.0

We can see that a new player will drastically win/lose points within his first games, while having a very low impact on the rating of established players. However, once a player reaches a rating>2100 his adaption speed is slowed down to normal in any case.