Forums / Community / Matchmaking Feedback & Discussion

[Locked] Matchmaking Feedback Update – May 21

OP ZaedynFel

  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 40
UPDATE:
The changes to the CSR update to bring it more in line with a player's MMR are now live. With the change, CSR will go up +15 on a win as long MMR is also going up (TrueSkill2 agrees with increasing skill). Players will also only lose -1 if their CSR is below their MMR when they lose. This way, players who end up with losses despite high performances will quickly regain the lost CSR, and keep their CSRs representative of their actual skill.

Likewise, players with overestimated CSRs will lose -15 on a loss as long as their MMR is below their CSR, but still gain +1 on a win. This prevents and undoes the negative effects of CSR inflation that tend to happen when a player is carried, or participates in a win even though they performed poorly.

The changes will slowly move everyone towards where they should be, but may not fully correct the inflation and deflation by the end of the season.

Next season, all players will place 200 CSR behind their MMRs instead of 100 because the new system converges much faster. In addition, no one will start above 1700 CSR.

This will result in next season having some of the most accurate visible Ranks we've had in H5 to date.

We know these changes will feel frustrating to players with inflated CSRs, but feel that the benefits of the ranks reflecting actual skill are important. They ensure you know how much you can still improve, and they also send the right message to your opponents at the end of a match. Currently, PGCRs are showing matches that look much less fair than they actually are. This will improve that greatly.

Two main comments this week.
Why am I matching against Champ / Onyx parties, often with only Diamond-level (or worse) teammates?

We have seen concerns with matching what appear to be stronger teams than expected. What you are seeing is their visible ranks, or CSR, and is not the most accurate measure of skill. MMR is way more accurate at predicting who will win. Usually this isn’t a problem because the two values stay relatively close to each other. There is a new issue where there are players with CSRs significantly higher than their MMRs. What appears to be a team of Champs is often actually a team of average Diamond players. Rest assured that despite appearances the matches are close in MMR.

To reduce the confusion, we will be rolling out a change in the coming weeks that will keep CSR more in sync with MMR. A player who with CSR above their MMR will not gain points on a win and will lose a small number of points on a loss. On the other hand, players with CSR below their MMR will not lose points on a loss and will gain a decent amount of points on a win. Players will see their CSR converge to their MMRs over the course of several matches.

Because the season didn’t begin with the change, not all players will have their CSRs in the correct spot by the end of the season, and while not perfect, this won’t affect who they get matched against.

Starting next season, players will start with their CSR behind their MMR like usual, Champs will start at 1700, and then everyone will climb towards their MMR. Once players reach their MMR, they will stay in sync after that. MMR can still change over the course of the season, but CSR will quickly converge on it.

So, in short, you are almost certainly not matching Champs with Diamonds for teammates, you are actually all Diamonds and we will make this more clear soon.


Clamping MMR for Matchmaking and Team Selection

As you may recall, we clamp MMR when it goes into the matchmaker to make sure players at the ends of the skill spectrum still find matches in a timely manner. For example, assume Diamond players are between 1 and 2, and Onyx is 2 and up. If we clamp at "2", then a 3.5 Onyx player becomes a 2.0 Onyx player, and looks not much better than a high Diamond player.

The actual clamping points are determined by the population size per playlist, aiming to guarantee at least a match worth of players will be available during any given search. In less popular playlists, the matchmaker can not distinguish between, e.g., Diamond and top Champ players when considering the set of players in a match.

There are two places the matchmaker uses the clamped values. They determine the overall top to bottom skill in a match, for example, whether you can have both Onyx and Platinum in the same match. They are also used in a team balance evaluation where the matchmaker makes sure that, given a potential set of players and their parties for a match, the team balancer will be able to create at worst a 40% chance for either team to win.

In the past, we used the same clamped value for both situations. This means that the team checker would allow matches that the skill system were unfair because it didn't know the values were clamped. We now use the raw and not clamped value for the team validation. Using the raw value won’t have any effect on most matches. It only influences the endpoints (very good and very bad players). But in those cases, it prevents situations where full parties and exceptional players with MMRs past the clamping locations can get into unfair matches.

For example, without the change, a full party of high-Onyx players appears to the matchmaker as low-Onyx solo players, resulting in the matchmaker treating a match between them and a group of low-Onyx solo players as totally balanced. It also would consider matching them with high-Diamond players as a 60/40 compromise. With the change, the same fireteam is now recognized by the matchmaker as both a full party and as high-Onyx and will match appropriately.


Playlist Popularity
No surprises. Keep in mind Grifball and Triple Team were only up part of the week.

Super Fiesta
Slayer
Warzone
Heroic Warzone Firefight
Castle Wars
Quick Play
Big Team Battle
Infection
Legendary Warzone Firefight
SWAT
Team Arena
Doubles
HCS 2018
Elimination
Grifball
Free-for-All
Mythic Warzone Firefight
Warzone Assault
Triple Team
Some really crunchy changes here, addressing concerns I've had with the system for a long time, namely unbalanced matches and inaccurate CSR. Well done!

Can't wait to have all these improvements form day 1 with Halo 6.
I dont understand how csr can get so inflated. If someone has a 90% win rate and tue skill 2 assesses that they arediamond, doesnt that just mean the original assessment was wrong? If they are truly a diamond and facing other diamonds shouldn't their win percentages be around 50%?

Also why are people getting placed into radically different skill levels on various smurfs. If the system is accurate shouldn't the same player place into the same place every time?
Also why are people getting placed into radically different skill levels on various smurfs. If the system is accurate shouldn't the same player place into the same place every time?
I placed Diamond 4 on my main account (4000+ games), and have since dropped to Diamond 1.

On my 2 alternate accounts, I placed Diamond 1, both times (less than 200 games, per account).

I'd say it's pretty accurate.
If I win a game I should move up regardless of how good your system thinks I am. I always loved the chase to improve my CSR in this game. The chase to get champion. It is fun to play and rank up. I am motivated to play ranked in order to get better and win games, if I consistently win and am communicating with my teammates then I should rank up.
I placed Diamond 4 on my main account (4000+ games), and have since dropped to Diamond 1.

On my 2 alternate accounts, I placed Diamond 1, both times (less than 200 games, per account).

I'd say it's pretty accurate.

My friend placed d5 on his main in dubs and champ 7 on a brand new smurf, I have placed anywhere from plat 3 to onyx 1800, from my perspective it seems like its not very good at predicting what rank you should be.
Can you please rephrase the whole clamping section or just give a low down on it? You lost me all throughout this section?

Also- does MMR have a cap? In other words- is there a certain point at which MMR is no longer calculated? A cut off- if you will...
ZaedynFel wrote:
To reduce the confusion, we will be rolling out a change in the coming weeks that will keep CSR more in sync with MMR.
I'm guessing it's going to create more confusion in the short term. You should probably put this thread in your sig so you can tell them to reference it, lol.
D M4N8 wrote:
Can you please rephrase the whole clamping section or just give a low down on it? You lost me all throughout this section?
Basically, ranks like champions and high onyx players for example will be able to match much lower ranked players so they can get a game, but the matchmaker will try to balance it so each team has a 40% chance of winning. His explanation was more technical, but that's what I got from it.
LUKEPOWA wrote:
ZaedynFel wrote:
To reduce the confusion, we will be rolling out a change in the coming weeks that will keep CSR more in sync with MMR.
I'm guessing it's going to create more confusion in the short term. You should probably put this thread in your sig so you can tell them to reference it, lol.
D M4N8 wrote:
Can you please rephrase the whole clamping section or just give a low down on it? You lost me all throughout this section?
Basically, ranks like champions and high onyx players for example will be able to match much lower ranked players so they can get a game, but the matchmaker will try to balance it so each team has a 40% chance of winning. His explanation was more technical, but that's what I got from it.
Sweet
Thanks.
Kind of seems like how it is now- though...
I've had so many close games with the new system, that I feel leaving it this way would be better.
D M4N8 wrote:
Can you please rephrase the whole clamping section or just give a low down on it? You lost me all throughout this section?

Also- does MMR have a cap? In other words- is there a certain point at which MMR is no longer calculated? A cut off- if you will...
It doesn't have a cut-off on the backend, but it does in some parts of the matchmaker. When comparing player to player, it cuts off according to a playlist's pop. When comparing team to team it has no cuttoff.
If I win a game I should move up regardless of how good your system thinks I am. I always loved the chase to improve my CSR in this game. The chase to get champion. It is fun to play and rank up. I am motivated to play ranked in order to get better and win games, if I consistently win and am communicating with my teammates then I should rank up.
That's not a skill system though, that's a grind / XP system and makes Ranks meaningless. If you want that, then grind 152.

The Ranks are supposed to be a measure of skill, not a measure of how many games you have played. We don't want worse players ranking higher just because they played more. That's not a skill-based system.
I placed Diamond 4 on my main account (4000+ games), and have since dropped to Diamond 1.

On my 2 alternate accounts, I placed Diamond 1, both times (less than 200 games, per account).

I'd say it's pretty accurate.

My friend placed d5 on his main in dubs and champ 7 on a brand new smurf, I have placed anywhere from plat 3 to onyx 1800, from my perspective it seems like its not very good at predicting what rank you should be.
Which gamertags and playlists? I haven't seen this. By far the majority get the same ranks on their alts.
I dont understand how csr can get so inflated. If someone has a 90% win rate and tue skill 2 assesses that they arediamond, doesnt that just mean the original assessment was wrong? If they are truly a diamond and facing other diamonds shouldn't their win percentages be around 50%?

Also why are people getting placed into radically different skill levels on various smurfs. If the system is accurate shouldn't the same player place into the same place every time?
Every case I've checked has shown smurfs in the same playlist placed almost exactly in the same place.

As for win rate, no, that's incorrect. Most players have near exactly the win rate TrueSkill2 expected, even the ones with 90%. The reason they have high win rates it the matchmaking, not the skill system. It means the playlist can't support their skill with its pop.
Is MMR calculated for each playlist or is it one raw score that aggregates your performance across all of Halo 5? Or is there a score for Arena and a score for Warzone?
NightClerk wrote:
Is MMR calculated for each playlist or is it one raw score that aggregates your performance across all of Halo 5? Or is there a score for Arena and a score for Warzone?
There is a separate MMR per playlist.
ZaedynFel wrote:
If I win a game I should move up regardless of how good your system thinks I am. I always loved the chase to improve my CSR in this game. The chase to get champion. It is fun to play and rank up. I am motivated to play ranked in order to get better and win games, if I consistently win and am communicating with my teammates then I should rank up.
That's not a skill system though, that's a grind / XP system and makes Ranks meaningless. If you want that, then grind 152.

The Ranks are supposed to be a measure of skill, not a measure of how many games you have played. We don't want worse players ranking higher just because they played more. That's not a skill-based system.
So when you release this new update in the following weeks, we will no longer gain CSR from wins, if our MMR isn't going up as well? Ie. If your MMR says you're a plat player, you won't be able to gain CSR and get into diamond until you have a diamond MMR as well?

I'm not sure how to feel about that, but hopefully it works out as intended.
ZaedynFel wrote:
If I win a game I should move up regardless of how good your system thinks I am. I always loved the chase to improve my CSR in this game. The chase to get champion. It is fun to play and rank up. I am motivated to play ranked in order to get better and win games, if I consistently win and am communicating with my teammates then I should rank up.
That's not a skill system though, that's a grind / XP system and makes Ranks meaningless. If you want that, then grind 152.

The Ranks are supposed to be a measure of skill, not a measure of how many games you have played. We don't want worse players ranking higher just because they played more. That's not a skill-based system.
So when you release this new update in the following weeks, we will no longer gain CSR from wins, if our MMR isn't going up as well? Ie. If your MMR says you're a plat player, you won't be able to gain CSR and get into diamond until you have a diamond MMR as well?

I'm not sure how to feel about that, but hopefully it works out as intended.
If your skill icreases to a "diamond" level then yes your csr will increase and you will rank diamond. All these changes are doing is just getting people's csr to match where their mmr is. So if your mmr is plat but you are ranked diamond your csr will drop until you are plat. It also works the other way though for people ranked below their mmr.

Now my only issue is that if this sort of system stays in place for the next year or halo 6 each playlist should have an xp rank like in halo 3 so that you can still be rewarded for grinding wins in a playlist. Rank 152 is not nearly enticing enough especially with how it is shared between all playlists in the game.
Also why are people getting placed into radically different skill levels on various smurfs. If the system is accurate shouldn't the same player place into the same place every time?
I placed Diamond 4 on my main account (4000+ games), and have since dropped to Diamond 1.

On my 2 alternate accounts, I placed Diamond 1, both times (less than 200 games, per account).

I'd say it's pretty accurate.
When you initially placed Diamond 4 was that with a full squad for all 10 placement matches? And did the alternate accounts have a mix of solo, to2, to3, &/or to4 during the placement matches? I ask because partnering up can inflate your visual skill level, but once you play solo or in fireteams that are less than full your MMR isn't as boosted. This means that your wins and losses should be occurring against lower ranked opponents than what you would have faced within a to4 and therefore outcomes can also directly effect your CSR placement and adjustments.
QX wrote:
ZaedynFel wrote:
If I win a game I should move up regardless of how good your system thinks I am. I always loved the chase to improve my CSR in this game. The chase to get champion. It is fun to play and rank up. I am motivated to play ranked in order to get better and win games, if I consistently win and am communicating with my teammates then I should rank up.
That's not a skill system though, that's a grind / XP system and makes Ranks meaningless. If you want that, then grind 152.

The Ranks are supposed to be a measure of skill, not a measure of how many games you have played. We don't want worse players ranking higher just because they played more. That's not a skill-based system.
So when you release this new update in the following weeks, we will no longer gain CSR from wins, if our MMR isn't going up as well? Ie. If your MMR says you're a plat player, you won't be able to gain CSR and get into diamond until you have a diamond MMR as well?

I'm not sure how to feel about that, but hopefully it works out as intended.
If your skill icreases to a "diamond" level then yes your csr will increase and you will rank diamond. All these changes are doing is just getting people's csr to match where their mmr is. So if your mmr is plat but you are ranked diamond your csr will drop until you are plat. It also works the other way though for people ranked below their mmr.

Now my only issue is that if this sort of system stays in place for the next year or halo 6 each playlist should have an xp rank like in halo 3 so that you can still be rewarded for grinding wins in a playlist. Rank 152 is not nearly enticing enough especially with how it is shared between all playlists in the game.
People are going to be salty lol. Now you really have to earn your rank. It isn't a bad thing obviously but winning games and not gaining csr in the process will take some getting used to.
  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 40