Halo 3 Forum
This topic has moved here: Subject: In depth explanation of the Halo 3 skill ranking system.
  • Subject: In depth explanation of the Halo 3 skill ranking system.
Subject: In depth explanation of the Halo 3 skill ranking system.
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

The simple proof is that a guy cannot get a 50 in 25 games playing with legit 10's at the beginning. Speed ranking only occurs with enablers.

  • 10.19.2009 8:58 AM PDT

"You build on failure. You use it as a stepping stone. Close the door on the past. You don't try to forget the mistakes, but you don't dwell on it. You don't let it have any of your energy, or any of your time, or any of your space." -Johnny Cash

Wow! That was a mouth full! But THX on explaining it in very specific detail!

  • 10.19.2009 9:03 AM PDT
  • gamertag: [none]
  • user homepage:

You haven't left yet?

what ive been wondering for a long time is does your rank increase outside of matchmaking gameplay? (i didn't read the whole thing so if thats on there i apoligize).

  • 10.19.2009 5:50 PM PDT

o_o o_0

finally i was wondering how it worked. i was getting very ticked but now that i know i understand and will not get as angry.

  • 10.20.2009 12:42 PM PDT
  •  | 
  • Honorable Legendary Member

Registration: SPARTAN II - 144
Birth ID: Aristotle
Code Name: Aegis Alpha, ALFA, Gold 4.
Appointed to: UNSC OpSec Command, Gold Team, [AA-S1c-444977]
Specs: Psy & Telep. abilities (ONI ref. #541-2285), Explosives, CQC & Tact., Sec. Protocols, Intel. Res. & Mangmt.

Posted by: kpicm
what ive been wondering for a long time is does your rank increase outside of matchmaking gameplay? (i didn't read the whole thing so if thats on there i apoligize).

If you mean outside of "ranked" matchmaking, then yes. There is basically an unseen "rank" for each Social playlist, since the system still tries to find appropriate matches for your particular skill/performance in that particular playlist.

Outside of matchmaking (ranked and unranked), there are no "ranks". The question becomes nonsensical, which is why I assume you meant "outside of ranked matchmaking" instead.

  • 10.20.2009 12:46 PM PDT

Wait, what? Uncertainty is lowered more for an unpredictable outcome compared to a predicted outcome? Also, the lower the sigma, the higher the MU gain? No, look again.Did you just ignore the graphs and formulas from the MS website I linked to? I'll go over it, step by step.

Let the unexpected outcome happen.
Lower teammate sigma -> Lower C.
muWinner<muLoser & Low C -> lower(muWinner-muLoser)/C
lower (muWinner-muLoser)/C + higher epsilon/C -> higher v(t,e)
Lower C2 -> higher sigma2/C2
higher sigma2/C2 & higher v(t,e) -> higher sigma2/C * v(t,e) -> Mu changes more.

More unexpected result -> uWinner-uLoser is smaller -> w(t,e) is higher -> 1-w(t,e) is lower -> final sigma is lower.

And if that for some reason isn't enough for you, go play around with the trueSkill calculator from the MS website and observe that every time you reduce the sigma of someone in an unexpected win, the others gain/lose more mu and lose more sigma. And when switching the result from expected to unexpected, sigma loss is increased.

In expected results, the Cs inside and outside the v() function go against each other, neutralising or reducting change when sigma reduces. But since the big gains are in unexpected results, getting the most out them is the fastest way to level up.

The primary point of losing is not to lose sigma, it is to lose mu. Because of the teammate's lower mu, the new player gets more credit for beating worse opponents. Also, the low mu makes them able to play the most legit low-level players that trueSkill would normally make them avoid due to the high starting mu (and as we know, "most legit"->low sigma-> lower enemy mu->lower). A new account isn't able to play the most legit players at their own mu, thanks to Bungie's 10 level search range.

The constant losing creates unexpected results until the quitter finally reaches the skill level that corresponds to a player who always quits. The sigma is changes as follows:
new sigma = old sigma*sqrt(1-(old sigma^2*w(t,e)/c^2)) + tau. When a player's sigma converges, the amount of sigma lost due to multiplication must be the same as tau. And unexpected results lead to a lower multiplier, which means that the 'convergent' sigma must also be lower so that the loss is equal to tau. The key here is that while the very final, converging sigma is the same for all levels and skills, the quitter is only on his way to his true level from a higher level. Because of this, he gets more unexpected results, which result in a lower sigma along the way. At the mentioned level 10, he still isn't as low as he will eventually end up at, and thus his sigma is lower than the sigma of a true legit level 10.

  • 10.21.2009 11:06 AM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

I got it and I am completely clueless at the same time.
I thought it was based on victories.

  • 10.25.2009 6:18 PM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

Campaign no for sure but Custom Games still remains a mystery.

  • 10.25.2009 6:19 PM PDT

Great job man. Very detailed, and informative. I have always been curious as to the basic structure of the ranking system. Nice Work!

[Edited on 10.27.2009 8:58 AM PDT]

  • 10.27.2009 8:57 AM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

Do you have a way to stop the negative Exp. boosters, because that gets on my nerves when you get a booster on the other team or on my own team and he has -2600 exp and quits befor the game is over.

[Edited on 10.27.2009 4:52 PM PDT]

  • 10.27.2009 4:51 PM PDT

thankyou very much for that detailed explination. is there seperate sigma and mu for every ranked playlist? or a total one for all ranked playlists?

anyway great job
i hope my 46 can get to a 50
he has 1090 exp
1000 ranked
800 social
reckon he can?

  • 10.28.2009 4:33 AM PDT

Posted by: Undisturbed Cow
Do you have a way to stop the negative Exp. boosters, because that gets on my nerves when you get a booster on the other team or on my own team and he has -2600 exp and quits befor the game is over.
They do ban people for habitual quitting but it does seem to take a while to "catch" them.

  • 10.28.2009 8:42 AM PDT

Cool.

To be honest as long as I just keep winning my games I should rank up every now and then.

But how do you explain this!

I lost three of my lone wolves games (6th place) in a row and then I ranked up.

  • 10.28.2009 9:12 AM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

yeah but in 2600 matches you would think thay would catch them.

  • 10.28.2009 2:39 PM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

To sum up system was made to lock people in to never rank up. Make new accounts ruin game for lower levles actually at 10-35 and make more money from 1 months or more.

  • 10.29.2009 9:49 PM PDT

The Song Of Nephilim

Xenoblade <3

the living dead systm is different I rank up 2 or 3 times sometimes but in all other playlists its only one.

  • 10.31.2009 4:25 PM PDT

Posted by: Mutoid Log
Wait, what? Uncertainty is lowered more for an unpredictable outcome compared to a predicted outcome? Also, the lower the sigma, the higher the MU gain? No, look again.Did you just ignore the graphs and formulas from the MS website I linked to? I'll go over it, step by step.

Let the unexpected outcome happen.
Lower teammate sigma -> Lower C.
muWinner<muLoser & Low C -> lower(muWinner-muLoser)/C
lower (muWinner-muLoser)/C + higher epsilon/C -> higher v(t,e)
Lower C2 -> higher sigma2/C2
higher sigma2/C2 & higher v(t,e) -> higher sigma2/C * v(t,e) -> Mu changes more.

More unexpected result -> uWinner-uLoser is smaller -> w(t,e) is higher -> 1-w(t,e) is lower -> final sigma is lower.

And if that for some reason isn't enough for you, go play around with the trueSkill calculator from the MS website and observe that every time you reduce the sigma of someone in an unexpected win, the others gain/lose more mu and lose more sigma. And when switching the result from expected to unexpected, sigma loss is increased.

In expected results, the Cs inside and outside the v() function go against each other, neutralising or reducting change when sigma reduces. But since the big gains are in unexpected results, getting the most out them is the fastest way to level up.

The primary point of losing is not to lose sigma, it is to lose mu. Because of the teammate's lower mu, the new player gets more credit for beating worse opponents. Also, the low mu makes them able to play the most legit low-level players that trueSkill would normally make them avoid due to the high starting mu (and as we know, "most legit"->low sigma-> lower enemy mu->lower). A new account isn't able to play the most legit players at their own mu, thanks to Bungie's 10 level search range.

The constant losing creates unexpected results until the quitter finally reaches the skill level that corresponds to a player who always quits. The sigma is changes as follows:
new sigma = old sigma*sqrt(1-(old sigma^2*w(t,e)/c^2)) + tau. When a player's sigma converges, the amount of sigma lost due to multiplication must be the same as tau. And unexpected results lead to a lower multiplier, which means that the 'convergent' sigma must also be lower so that the loss is equal to tau. The key here is that while the very final, converging sigma is the same for all levels and skills, the quitter is only on his way to his true level from a higher level. Because of this, he gets more unexpected results, which result in a lower sigma along the way. At the mentioned level 10, he still isn't as low as he will eventually end up at, and thus his sigma is lower than the sigma of a true legit level 10.


al leuaulualeulaue- What!?!?!

  • 10.31.2009 10:46 PM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

wow, thanks heaps!!

  • 11.01.2009 1:31 AM PDT

YouTube.com/KiesterKicker

Twitter.com/KiesterKicker

Very informative and straight to the point. Good job.

  • 11.01.2009 7:46 AM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

Just a quick question!

Do the different playlists effect eachother?

ex.

Player got 1000 exp and low sigma in lone wolves and then starts playing a new playlist. Will his skill, experience and previously played matches effect his skillgain?

I'm certain it has been asked before, but couldn't find any answer. :p

  • 11.04.2009 9:49 AM PDT

Together/dumped: 3-24-07 - 1-28-10
You said you'd never leave.. yet you walked away like im nothing.
A wise man once told me.. "Never give up on someone you can't go a day without thinking about.."
I will always miss you Zoe. Zoe is, was, and always will be the love of my life. I'll love you forever and wait for you to one day return to me and pick up the pieces of my broken heart. I Love You..

Good job.

  • 11.07.2009 11:59 PM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

Posted by: En rosa kanin
Just a quick question!

Do the different playlists effect eachother?

ex.

Player got 1000 exp and low sigma in lone wolves and then starts playing a new playlist. Will his skill, experience and previously played matches effect his skillgain?

I'm certain it has been asked before, but couldn't find any answer. :p


No

  • 11.08.2009 11:14 AM PDT
  • gamertag:
  • user homepage:
  • last post: 01.01.0001 12:00 AM PDT

Posted by: Nephilim713
the living dead systm is different I rank up 2 or 3 times sometimes but in all other playlists its only one.


You rank up quicker in FFA than team games...the system needs less games to figure out your skill level.

  • 11.08.2009 11:15 AM PDT

-----------------------------------------------------------------
Scott Allen

Woah - thanks for posting this, now we can see how flawed the Halo ranking system is. (Y)

  • 11.15.2009 9:36 AM PDT

I'm level 49. I've played about 70 games and have 48 exp i believe. I've been hoverring between 48 and 49 the last 30 games. At 49, I win 1 and stay at 49, lose 1 stay at 49, lose another move down to 48. Win 1, 49 again. Any way to tell how many wins it would take straight to get from 49 to 50? I'm at 49 again. Im figuring 2-3. jw

  • 11.15.2009 9:51 PM PDT