How rank placements are calculated

Hey so lately I’ve seen alot of posts about ranking placements and hopefully I can clear up how placements work through the best of my knowledge, personal experience, and testing with multiple playlists.

First, all players have a hidden “skill rating” or elo. this determines who you play against and what CSR you will be placed in. The ranking system is complicated and has too many factors for me to know entirely but “skill rating” should equal your CSR and rankings should be between a set amount of CSR points. Onyx is 1500 CSR as an example, and diamond could be something like 1100-1499. (this is an assumption)

Placement matches take into account SEVERAL things, and were more finely tuned for this season to allow placement into things other than div 1.
The things that can affect your placement are:
KDA, when you have a yellow tick on your score it adds extra points to your placement, having all 3 makes does even more.
Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.
Your last season CSR will affect the level of player you play against this season to start off with as the system uses this to match you against equal CSR. The system also uses your CSR in other playlists to match you, these are much more flexible as they are just estimating your skill.

Examples:
If you in 10-0 and you were plat 1 last season ending you will be placed higher. More than likely Diamond 2-5.
If you go 5-5 and were Plat 3 last season, welcome back to plat 3.
If you are bronze 1 and go 10-0 you can place Bronze 5 - Silver 6.

These are just estimates from knowledge and samples I have seen.

I was onyx 1700-1800, I won 8 out of 10 placement matches, the first 8 matches went perfect, even losing the last two my k/d wasn’t below 0.8. i was still placed in platinum.

> 2533274884665671;2:
> I was onyx 1700-1800, I won 8 out of 10 placement matches, the first 8 matches went perfect, even losing the last two my k/d wasn’t below 0.8. i was still placed in platinum.

Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.

this is probably what happened.

> 2533274884665671;2:
> I was onyx 1700-1800, I won 8 out of 10 placement matches, the first 8 matches went perfect, even losing the last two my k/d wasn’t below 0.8. i was still placed in platinum.

What gametype was it? If swat or FFA those were really really really broken and too many got rank in Onyx

> 2533274803518335;1:
> Hey so lately I’ve seen alot of posts about ranking placements and hopefully I can clear up how placements work through the best of my knowledge, personal experience, and testing with multiple playlists.
> First, all players have a hidden “skill rating” or elo. this determines who you play against and what CSR you will be placed in. The ranking system is complicated and has too many factors for me to know entirely but “skill rating” should equal your CSR and rankings should be between a set amount of CSR points. Onyx is 1500 CSR as an example, and diamond could be something like 1100-1499. (this is an assumption)
> Placement matches take into account SEVERAL things, and were more finely tuned for this season to allow placement into things other than div 1.
> The things that can affect your placement are:
> KDA, when you have a yellow tick on your score it adds extra points to your placement, having all 3 makes does even more.
> Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.
> Your last season CSR will affect the level of player you play against this season to start off with as the system uses this to match you against equal CSR. The system also uses your CSR in other playlists to match you, these are much more flexible as they are just estimating your skill.
> Examples:
> If you in 10-0 and you were plat 1 last season ending you will be placed higher. More than likely Diamond 2-5.
> If you go 5-5 and were Plat 3 last season, welcome back to plat 3.
> If you are bronze 1 and go 10-0 you can place Bronze 5 - Silver 6.
> These are just estimates from knowledge and samples I have seen.

Citation needed.

Seriously, this is a bunch of information which would be great if you could offer an actual source for it.

> 2533274803518335;3:
> > 2533274884665671;2:
> > I was onyx 1700-1800, I won 8 out of 10 placement matches, the first 8 matches went perfect, even losing the last two my k/d wasn’t below 0.8. i was still placed in platinum.
>
>
> Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.
>
> this is probably what happened.

why would it decrease, if I win against someone lower than me, that seems like it would be an issue with the searching system putting me against people lower. the two matches I lost we were down a player and it was against a mix of high platinum’s and one diamond. I think when they tweaked swat and ffa placement, I think they went a bit too much.

I have said and continue to say that the ranking system has got to be one of the worst ever. A single individual should not have to depend on other randoms to get to rank up or down. I fully understand that win/loss should matter to a certain degree. But not to the amount it currently does. After all if we are properly matched most peoples win/loss ratio should be near 1 plus or minus a tenth. It is my opinion that rank should consist of K/D + W/L + XP +Gun accuracy + Average Grenade kills per game. If properly matched, the W/L and K/D would level out over time and XP would eventually hit max level. However gun Accuracy and average grenade kills will still separate the good from the bad.

> 2533274872647908;4:
> > 2533274884665671;2:
> > I was onyx 1700-1800, I won 8 out of 10 placement matches, the first 8 matches went perfect, even losing the last two my k/d wasn’t below 0.8. i was still placed in platinum.
>
>
> What gametype was it? If swat or FFA those were really really really broken and too many got rank in Onyx

I was onyx in both swat and team arena and diamond in BTB and slayer. I didn’t do bad in onyx, It felt appropriate in team arena and swat. I’ve played against champions before and done fine.

> 2533274825101441;7:
> I have said and continue to say that the ranking system has got to be one of the worst ever. A single individual should not have to depend on other randoms to get to rank up or down. I fully understand that win/loss should matter to a certain degree. But not to the amount it currently does. After all if we are properly matched most peoples win/loss ratio should be near 1 plus or minus a tenth. It is my opinion that rank should consist of K/D + W/L + XP +Gun accuracy + Average Grenade kills per game. If properly matched, the W/L and K/D would level out over time and XP would eventually hit max level. However gun Accuracy and average grenade kills will still separate the good from the bad.

Sorry, but no. The problem of assigning a skill rating to a player is much more complicated than you think. First of all, the ratings are based on wins, but that doesn’t mean they are based on win-loss ratio. In short, the way win-based rating systems rate players is that they assign players a skill rating. When two players meet, their expected win probabilities are calculated based on the difference of their skill ratings. The expected win probability is at the heart of the system, because the goal is to optimize the win probability estimate in such a way that it corresponds as closely to the actual performance as possible. I.e. if we denote the estimate with E, and the actual performance with P, then E - P should be as close to zero as possible. Say, the player wins seven out of ten games against another player, then P is 0.7. If the estimated performance against that player was 0.8, we’d get 0.1 for E - P.

This is the basis on which all the successful rating systems work. The method is incredibly robust, because the changes in rating are roughly proportional to the size of E - P. This means that if the players actual performance doesn’t match their expected performance, their rating will change. This means that the only stable rating a player can obtain is the one which corresponds to their actual performance.

The problem you run into when you start combining all kinds of silly statistics is how do you count them in? At the end of the day, whatever system you devise, it’s expected to be an estimate of the player’s performance. if you pit two players against each other, how do you derive the win probability estimate? Does your system lead to stable ratings? If so, can you show that these stable ratings correspond to the player’s actual performance? If your system has no predictive power, it’s as good as useless.

> 2533274825830455;5:
> > 2533274803518335;1:
> > Hey so lately I’ve seen alot of posts about ranking placements and hopefully I can clear up how placements work through the best of my knowledge, personal experience, and testing with multiple playlists.
> > First, all players have a hidden “skill rating” or elo. this determines who you play against and what CSR you will be placed in. The ranking system is complicated and has too many factors for me to know entirely but “skill rating” should equal your CSR and rankings should be between a set amount of CSR points. Onyx is 1500 CSR as an example, and diamond could be something like 1100-1499. (this is an assumption)
> > Placement matches take into account SEVERAL things, and were more finely tuned for this season to allow placement into things other than div 1.
> > The things that can affect your placement are:
> > KDA, when you have a yellow tick on your score it adds extra points to your placement, having all 3 makes does even more.
> > Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.
> > Your last season CSR will affect the level of player you play against this season to start off with as the system uses this to match you against equal CSR. The system also uses your CSR in other playlists to match you, these are much more flexible as they are just estimating your skill.
> > Examples:
> > If you in 10-0 and you were plat 1 last season ending you will be placed higher. More than likely Diamond 2-5.
> > If you go 5-5 and were Plat 3 last season, welcome back to plat 3.
> > If you are bronze 1 and go 10-0 you can place Bronze 5 - Silver 6.
> > These are just estimates from knowledge and samples I have seen.
>
>
> Citation needed.
>
> Seriously, this is a bunch of information which would be great if you could offer an actual source for it.

There are no sources for this information as there are none out there, it is based on information I have gathered from the forums, friends, and personal experience. I was doing my best to explain the ranking system for the multitude of posts made in regards to it.

> 2533274825830455;9:
> > 2533274825101441;7:
> > I have said and continue to say that the ranking system has got to be one of the worst ever. A single individual should not have to depend on other randoms to get to rank up or down. I fully understand that win/loss should matter to a certain degree. But not to the amount it currently does. After all if we are properly matched most peoples win/loss ratio should be near 1 plus or minus a tenth. It is my opinion that rank should consist of K/D + W/L + XP +Gun accuracy + Average Grenade kills per game. If properly matched, the W/L and K/D would level out over time and XP would eventually hit max level. However gun Accuracy and average grenade kills will still separate the good from the bad.
>
>
> Sorry, but no. The problem of assigning a skill rating to a player is much more complicated than you think. First of all, the ratings are based on wins, but that doesn’t mean they are based on win-loss ratio. In short, the way win-based rating systems rate players is that they assign players a skill rating. When two players meet, their expected win probabilities are calculated based on the difference of their skill ratings. The expected win probability is at the heart of the system, because the goal is to optimize the win probability estimate in such a way that it corresponds as closely to the actual performance as possible. I.e. if we denote the estimate with E, and the actual performance with P, then E - P should be as close to zero as possible. Say, the player wins seven out of ten games against another player, then P is 0.7. If the estimated performance against that player was 0.8, we’d get 0.1 for E - P.
>
> This is the basis on which all the successful rating systems work. The method is incredibly robust, because the changes in rating are roughly proportional to the size of E - P. This means that if the players actual performance doesn’t match their expected performance, their rating will change. This means that the only stable rating a player can obtain is the one which corresponds to their actual performance.
>
> The problem you run into when you start combining all kinds of silly statistics is how do you count them in? At the end of the day, whatever system you devise, it’s expected to be an estimate of the player’s performance. if you pit two players against each other, how do you derive the win probability estimate? Does your system lead to stable ratings? If so, can you show that these stable ratings correspond to the player’s actual performance? If your system has no predictive power, it’s as good as useless.

Thank you. That actually makes a lot of sense. Pretty much confirming most of what the OP said.

> 2533274803518335;10:
> There are no sources for this information as there are none out there, it is based on information I have gathered from the forums, friends, and personal experience. I was doing my best to explain the ranking system for the multitude of posts made in regards to it.

You should maybe stress more that your OP is purely speculation then. But as this is just your speculation, there are few things I have to criticize.

> 2533274803518335;1:
> First, all players have a hidden “skill rating” or elo.

I saw you saying TrueSkill in in another thread, so which is it? Elo ratings are not synonymous with skill ratings, you known. And considering 343i’s affiliation with Microsoft, it’s far more likely that they are working with TrueSkill as the back bone rather than Elo ratings. And the difference is kind of significant because Elo only works with one variable while TrueSkill works with two.

> 2533274803518335;1:
> The ranking system is complicated and has too many factors for me to know entirely but “skill rating” should equal your CSR and rankings should be between a set amount of CSR points.

Oh, not at all. Elo system has a fairly good Wikipedia page and is really not that complicated to explain the basics of, and TrueSkill, although uncomfortably complicated once you go down the rabbit hole, is still simple to explain in principle as long as you don’t go into great detail about team-based matchmaking and rating updates.

Elo I almost explained in my previous post. The system uses differences in players’ skill ratings to compute their win probabilities (expected performance). It then compares the expected performance to the actual performance, and the resulting change in rating is proportional to that difference. If the real performance is more than the expected, the rating goes up, and if it’s less, the rating goes down. The rating change is also proportional to a number called the K-factor, by which the performance difference is multiplied to arrive at the final change in rating.

The K-factor is not that significant when we discuss Elo ratings, but matters more when we discuss TrueSkill. The short story is that TrueSkill is “like Elo” (notice the quotation marks), but with a variable K-factor which also changes based on how much the player’s performance corresponds to their expected performance. The reality is of course much more complicated. The update formulas of TrueSkill are much more complicated than those of Elo, and not really worth discussing because of that. But the gist of it is that aside from just keeping track of player skill, TrueSkill also keeps track of how reliable the estimate is, which makes it much more flexible.

> 2533274803518335;1:
> Placement matches take into account SEVERAL things, and were more finely tuned for this season to allow placement into things other than div 1.
> The things that can affect your placement are:
> KDA, when you have a yellow tick on your score it adds extra points to your placement, having all 3 makes does even more.
> Your W/L matters the most heavily with extra weight on people either higher or lower than your CSR. If you win against someone higher than your CSR your placement will increase, and lower will decrease your CSR.
> Your last season CSR will affect the level of player you play against this season to start off with as the system uses this to match you against equal CSR. The system also uses your CSR in other playlists to match you, these are much more flexible as they are just estimating your skill.

It’s unlikely that K/D has any effect on ratings. 343i were very clear that winning is the only way to affect your rating. The last season’s CSR affecting placement matches I could get behind though. I’ve said before that throwing away a completely good skill estimate is just silly. This is also where the advantage of TrueSkill comes in: blowing up the rating uncertainty to some big value at the start of every season while retaining the old skill estimate is much more sensible. It’s in essence as if your rating had been wiped clean, but it still tries to match you with players of appropriate skill level from the get go.

Same story for me. I did really good the first 7 matches did ok on the next 2 then bombed the last one and got a Gold 6 when I was an Onyx (1600-1700) last season. So stupid. And I won 7 out of 10 matches so…

Can you keep yor rank fi losing?

> 2535411123013729;14:
> Can you keep yor rank fi losing?

Please don’t revive old topics thanks and No. No you can’t.