Example
https://www.xbox.com/play/media/CHWNDTJ7
Got destroyed in this game because a player quit. So I did nothing but try to take out a couple of their guys out.
Lost 4-0.
Didn’t lose any rank
Fully played the objective this game.
.Won 4-1
Rank only went up a tiny tiny bit.
https://www.xbox.com/play/media/YER8A3A6
It’s still based pretty much on how many people you kill. That’s it.
How much CSR you change depends on the result, the relative strength of your opponents, and the pull off your MMR (if they are wide apart).
If you want to rank up you have to beat a team ranked above you. And to rank down, lose to a team ranked below you. Other results will have minimal change.
There is an adjustment if you lose players early (it seems to be limited to +/- three points).
Kills don’t seem to be a huge factor.
I’m actually running a spreadsheet. I have over 300 data samples now. And I’m looking for correlation to CSR change.
So far I have found no correlation to kills (correlation score of 0.28).
Similarly results for deaths and damage.
There is “poor” correlation with K/D, KDA, K rate, total score, and objective score.
Interesting that K/D and Objective score have very similar correlation (0.45 and O.36). People swear black and blue that it’s better to “slay” than “play” (the objective). But this is clearly not true.
Anecdotally I’ve been wondering if K/min has an increased weighting during placement. But it’s clearly not an ongoing issue.
I have no idea what you are saying in most of that but, are you saying kills aren’t a huge factor?
Probably not.
And definitely not any where near as much as some people make out.
The TrueSkill2 discussion paper specifically talks about kills and how they don’t add any extra information over the win itself. Which makes sense. If you get a lot of kills there is a good chance you won. At least in Slayer.
Kills per minute is a different metric. But the use of that is identifying a big fish in a small pond and ranking them up faster (not higher). Imagine an Onyx in game vs Silver and Golds. They are going to get a lot of kills in a short period of time. But if you are playing against equals - kill rates are very hard to bump. You can manipulate K/D. But not K/min. It’s an independent measure of skill.
What I’ve done in my spreadsheet is collect all the CSR changes and chart them against K, D, A, K/D, etc. I then use the CORREL function to look for correlation. After 300+ data points the correlation between kills and CSR change is 0.28. It needs to be at least 0.7 to show a strong correlation.
So yes. I am saying that kills aren’t a significant factor. People tend to notice those times that someone with a bigger K tally get more CSR and get upset. Especially when they played the objective.
But the extra CSR is more to do with the shape of their MMR curve.
Objective or slayer, the only factors your CSR increases by are Win/ Loss and Positivity of your KD.
Nothing else matters.
It kind a does cause winning the game helps you go up and to win the game you have to play the objective.
It’s just a shame that getting 30 kills a game helps you go up more. Maybe they should just give you less points for getting kills.
Unfortunately, it doesnt matter if its a ranked slayer or objective match. K/D is has much more weigth the holding the ball, bringing back the flag and so on.
And thats just wrong.
In a ranked objective match, kills should be rewarded with only 10% of the points that you get in a ranked slayer match.
It has nothing to do with kills. It has to do with damage.
So let’s say you got to Onyx by playing objective and never getting kills.
Your teammates would be carrying you, because you can’t contribute via damage.
You get put in a higher skill level when you do enough damage / win consistently. Because it shows you’re contributing to the fight. But if you’re just like, only playing OBJ, you… well, don’t belong in a higher grade. You’ll just get creamed.
Shoot lots of bullets but make sure you hit…!
Over 300 data points in my spreadsheet.
K/D has low correlation to CSR increase in a win (r = 0.40) and no correlation in a loss (r = 0.22).
This suggests that the win and the ranking of your opponents (pull of your MMR) are what matter the most.
For a win the correlation with CSR change is 0.44 for raw kills, 0.40 for KD, 0.42 for KDA, and 0.47 for K/min.
Low correlation across the board. So if it is a factor it is lost in the noise of MMR influences (ie. the team averages).
The correlation for Total Score is 0.39 and Objective Score only 0.11. This would suggest that while kills are more important, neither are strong enough to significantly affect your CSR.
See above.
I would love it if they did more with damage stats. I think they have the potential to tell way more of the story than simple K and D.
But neither Damage (r = 0.33) or Damage per minute (r =0.28) seem to hold any influence.
I’m going to need more stats. And over a bit longer of the season. Early on the pull of MMR is much stronger (everyone’s CSR is lower than their MMR). Late in the season the MMRs will tend to rank lock. I reckon we’re kind of in the sweet spot now where trends should pop out?
And just in case anyone is interested. Now that we can see CSR changes on a per game basis;
The most I have seen in any game is plus or minus 15 CSR points. This goes against the old fashioned claim of “losing over half a bar”.
Quitters lose -15 points.
The token CSR gain when the system doesn’t have enough info to adjust MMR seems to be +/- 3 points.
In one game I had three people quit early and the last person quit after about 4 minutes of running around 1v4. Only one of the early quitters showed in the post game stats and they lost -15 points. The late quitter only lost -3 points. All the those on the “winning” team got +3 points.
I’d like if we can see it no matter the gametype tbh.
Dunno’. The pros swear by it. Maybe I can pull over @Liquid_Execute for this one. Hey buddy, what’s your input here?
Kills and damage both matter.
It can feel like you get “rank locked” but moreso the game just catapulted to generally what your rank should be.
As annoying that is, the game pulls you to where your MMR is basically no matter what. You are in effect grinding to make your MMR higher, not your CSR.
Cool. I’m going to change my style of play now. See how it goes.
What I would love to see is a break down of damage into effective (leading to a kill) and escaped.
You could really go deep into people’s playing styles and form.
And imagine all the post game medals you could derive from the various combinations!
Cool insight.
Do you have an opinion in regards to raw K vs K/D vs KDA vs K/min?
The original TrueSkill2 discussion paper precedes the use of damage stats. And it definitely makes sense to use them. Even more than kills. And I guess total damage is really just a more accurate version of KDA.
It would be awesome to know the various metrics are weighted - but I can see why 343 don’t really want to advertise them.
My anecdotal suspicion is that they have more bearing early (placement). People seem to be able to retain their previous season’s MMR on the back of personal performance over despite not winning.
At the moment they seem to be lost in background to the pull of your MMR on the CSR.
And I guess the important insight we need from a very high level player is - do you think the kills and damage outweigh the objective. Are you ever holding the Oddball and thinking to yourself that this is bad for my MMR?
The magic of TrueSkill2 
Isn’t this a good thing though?
As long as your MMR is the more accurate representation of “skill”.
Definitely not how the system works… I get matched probably 9/10 with the other team boosting Onyx vs low plats and Diamond somehow the mmr at the end of the match said we were close… not even close. How does a team with 2 onyx players a D6 and gold have the same mmr as plat 5, plat 6, D1 and D3 lol this ranking system is so broken… once you are onyx it’s pretty impossible to lose it even if you lose and go negative 15
You have to remember the CSR increments are going to be low as they are scaled down from the ratings (MMR) that is scored in the game. If this wasn’t the case CSR would lose its purpose of an orderly progression/regression system that the player can follow.
You can’t see MMR move up and down in Ranked but thanks to Halo data hive it shows a good visual example in FFA. Like we said before I don’t think it would be best to follow LSS as it seems to be working in a different scale to FFA and Ranked due to the nature of the game and how points are scored in your MMR.
But what you can see in the sporadic evolution of events Halo Data Hive shows is that Damage is one of the biggest factors along with wins that will push your MMR higher.
Unfortunately with your research this will probably not become more apparent until you have a lot more results. Due to the lower/slower increments CSR moves at.
Interesting what you have though. You should post it once you are finished with your sample so people can comment on your findings.
I had a game like this the other week. Guess who got the most kills - the gold player.
In defence of the match making system it has matched together an overall team average. Which would equate to about 1300 I’m guessing just looking at your line up of players.
But you could argue an Onyx player shouldn’t be allowed to go into ranked with anybody less than a Platinum.
But they are constantly moving in the direction of your MMR.
And I realise that my stats are limited in what they can tell me about how the MMR moves. That is true. But my premise (or null hypothesis) is that you can’t use kills or k/d etc to predict your CSR change.
People can’t complain that “I won and had the best K/D but my CSR hardly moved”.
I think it’s dangerous to try and use FFA to judge MMR behaviour. Do we even know how they allocated winners and losers? Is it just one winner and 5 losers? Are you ranked each time as a loser vs everyone who finished above you AND as a winner against everyone who finished above you?
The latter option kind of makes sense - but could cause some serious MMR fluctuations. Especially early with a relatively volatile MMR curve.
It would take some serious time to collate the data. You’d have to play a game. Get your MMR and then collate it against damage but also all the other stats to show that damage was actually a stronger influence.
And don’t get me wrong. I love damage as a stat. It has so much more potential than simple K and D. It could be a powerful ranking tool.
I’ll give it some thought but it would be considerably more time consuming - and you’d have to play a lot of rumble pit.
Don’t get me wrong. I’m not discounting the potential for any game to be unbalanced.
But you can’t really use the CSR to estimate the average MMR.
The CSR is a conservative representation of skill. All we can say is that there is 98% chance the MMR is higher. But without knowing the width of the MMR curve we don’t know how much higher.
For the Onyx players it’s going to close. Probably nigh on identical.
But your Platinum’s could easily have a Diamond MMR. And your Diamonds, Onxy.
And yes, I too am getting a bit sick of all the Onyx players sandbagging with lower ranked players to help them grind rank. A couple of games ago I came up against a Onyx/Onyx/Bronze/Unranked combination. Seriously?
The only consolation was that the latter two were really bad and we managed to win the game. Despite all four of us being Platinum. And me having to deal with 200+ms lag.