well this isnt exactly right, and theres a lot more factors that affect it (decay, weapon weights, bonuses, etc).... but heres an outdated explanation
Quote: Originally Posted by psychostats.com
Alternative Psychostats 3 ranking
The alternative ranking algorithm seems to be much better in ranking top-players. Problem
with the default ranking algorithm is that even players who play against complete noobs can
get their rank very high. The new system does not work that way. If someone wants to be in
top-10 he has to play against people in the top-10.
Algorithm for computing how skill changes according to kills and deaths is bit more
complicated than the default.
# Pro kills noob (killer's skill is higher than the victim's)
1.1 Pro gets points: (kskill + vskill)^2 / kskill^2
1.2 Noob loses points: ((kskill + vskill)^2 / kskill^2) * (vskill / (vskill + kskill))
# Noob kills pro (killer's skill is lower than the victim's)
2.1 Noob gets points: (vskill + kskill)^2 / vskill^2 * vskill/kskill
2.2 Pro loses points: ((vskill + kskill)^2 / vskill^2 * vskill/kskill) * (vskill +
baseskill) / (vskill + kskill)
# For example
# Pro (A=1000) kills noob (B=200)
# Pro gets points by formula (1.1)
(1000 + 200)^2 / 1000^2 = 1.44
# Noob loses points by formula (1.2)
((1000 + 200)^2 / 1000^2) * (200 / (200 + 1000)) = -0.24
# Noob (B=200) kills pro (A=1000)
# Noob gets points by formula (2.1)
(1000 + 200)^2 / 200^2 * 200/1000 = 7.2
# Pro loses points by formula (2.2)
((1000 + 200)^2 / 200^2 * 200/1000) * (1000 + 50) / (1000 + 200) = -6.3
# When pro player (A=1200) kills another pro player (B=1100)
# Killer (A) gets points by formula 1.1 = 3.67 points
# Victim (B) loses points by formula 1.2 = -1.76 points
The goal of new ranking system was to have smaller weight for players who kill mostly
noobs.
Each player starts with a baseskill of 50. Someone who has previously succeeded well in
ranking by killing noobs would now have to kill around four 50-skill-players to achieve
same skill increase as killing one another good player.
its definitely not perfect in any sense at all.. but its better then the old ELO method