Forums / Community / Matchmaking Feedback & Discussion

[Locked] Matchmaking Feedback Update – October 9

OP ZaedynFel

  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 4
Team Fairness Update

Last week we rolled out a new team matchmaking rule that does a much better job at making sure a match is fair when parties are involved. The early results were promising, but we discovered an issue that forced us to take the rule down for the time being. The issue doesn’t sit on the 343 side of the fence, so I don’t have an ETA now.


12-Player Fireteams Warzone

Being able to allow 12-players in Warzone was dependent on the rule above. Without it, we can’t efficiently control team vs. team imbalances, so we again reduced the max Fireteam size to 6

Feedback had been mostly positive, with a few of you pointing out that it was still possible to get into one-sided matches. We did see matches that look unfair when looking at the raw skill gap between teams, however the matches would have looked fair to the matchmaker because it treats players above a certain skill level as equal.

If we run 12-player Fireteams again, we will probably move that cut-off point up higher. This will mean that the best 12-player Fireteams won’t default to playing weaker amalgamations of solo players and will instead wait indefinitely for other top teams to play.

I won’t have further updates on this feature until the rule discussed above is re-introduced first.
Let’s talk about this new matching algo for to4 in arena. Doesn’t it undermine the gold, plat, diamond, etc ranking system? If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against 4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by 4 random champs.
Can you please explain to me why you insist on catering to solo searching players in team playlists? The point of the team playlist is communicate and play like a team. If you can’t find 12, there is 8, if you can’t find 8 there is 4, if you can’t find 2, there is FFA. People who don’t want to communicate and don’t want to party up should expect to feel some pain. The people who are smart enough to play with a team in a team playlist should not be penalized because potatoes want to be potatoes. This applies to both arena opponents and wz teammates.
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when data suggests that expected performance adjustment are warranted based on how certain players play together when in a party. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.

Also, halo 5 is probably 95% solo players or small fireteams (Josh probably has the exact numbers), especially in social modes. So I appreciate that they are trying to come up with a MM system that works acceptably well both for parties and for solo players.
ZaedynFel wrote:
12-Player Fireteams Warzone Feedback had been mostly positive, with a few of you pointing out that it was still possible to get into one-sided matches. We did see matches that look unfair when looking at the raw skill gap between teams, however the matches would have looked fair to the matchmaker because it treats players above a certain skill level as equal.
If we run 12-player Fireteams again, we will probably move that cut-off point up higher. This will mean that the best 12-player Fireteams won’t default to playing weaker amalgamations of solo players and will instead wait indefinitely for other top teams to play.
I won’t have further updates on this feature until the rule discussed above is re-introduced first.
This is great news. I think allowing 12-man has a lot of potential for great matchups at the high-skill tier, and ALSO allows people of lower skills to still party up and play with 11 friends, but get put into fair matches against higher-skill solo players rather than waiting hours to finally match a To12 god-squad. But it just has to be done right... allowing 2.0 MMR solo players (or small fireteams) to constantly match a god squad To12s with all 4.0 MMRs is just not a good option.

I'd like to add - I think it's really important to lower the restrictions on player-to-player skill, but keep the team-to-team skill tight. Many folks have been complaining that the recent MMR tightening has made it virtually impossible for high MMR players to find games in any social playlist. While it's fair for TEAMS with insanely high MMR to have matchmaking struggles in a social playlist (because no other team has comparable team MMR), those high MMR players should be able to search solo (or with a small fireteam) and still find games. It's not right for people who paid money for the game, and pay an XBL subscription, to be excluded from playing at all.
RzR J3ST3R wrote:
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.
ZaedynFel threw up a good post to close the earlier thread on this where he noted the idea is to take the odds of full team winning against groups of solos from like 99:1 to 3:1. It's worth a read. Sounds like lots of game data had been looked at to determine where things need adjusting. I mean....99:1 isn't a game....it's shooting fish in a barrel.
One other suggestion - I think this basically already happens, but it would be good if selecting "expanded" search would allow you to match players and teams of more varied MMR, meaning you'd tolerate matches against significantly better or significantly worse players and/or teams, but only if those other players can't find better matches elsewhere! That way if you're searching in a To12, you have the best possible chance of matching another To12 searching at that time. Otherwise it seems possible for 2 teams (with somewhat different MMRs) to TRY to match each other, but still both fail to find any game at all, because their MMRs are too far apart.

On the other hand, if moderate-skill players want to avoid matches with blowout potential, they just search focused or balanced and the MM will be tighter, though they may wait a lot longer.

But it just seems so silly that I solo-search a social playlist like WZA on "expanded" and can't find games at all...
Thank you for your effort to bring back larger fireteams for warzone.
Take the time you need to get it right. Please just bring it back when it’s ready!
We were already matching the same few lobbies now it's going to be even tighter, I don't think we're going to be able to get matches if 12 man is re released then.
RzR J3ST3R wrote:
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.
ZaedynFel threw up a good post to close the earlier thread on this where he noted the idea is to take the odds of full team winning against groups of solos from like 99:1 to 3:1. It's worth a read. Sounds like lots of game data had been looked at to determine where things need adjusting. I mean....99:1 isn't a game....it's shooting fish in a barrel.
Right, like has been said, it's not an adhoc guess at how much better a team should be.

It's learned using a cutting edge statistical machine learning method that looks through 100s of millions of matches and learns, by playmode and party size, what that advantage is.

We've already evaluated the prototype on real matches and found it's very accurate at predicting the win probability. For example, it's around a half a tier for a lot of our Ranked 4v4 modes.

So we found, in our data, a mid-tier to4 Diamonds wins half their matches against low-Onyx solos.
How does matchmaking work for 343 unicorn skin play dates? Are skill and fire team size restrictions removed so everyone has a fair chance?
some lag and the teams arent very even. and the reqs are rigged. if you use a victory boost 80% chance you will loose.
some lag and the teams arent very even. and the reqs are rigged. if you use a victory boost 80% chance you will loose.
more figherfight maps are needed. and multiplayer maps. more loadout options too.
RzR J3ST3R wrote:
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.

Also, halo 5 is probably 95% solo players or small fireteams (Josh probably has the exact numbers), especially in social modes. So I appreciate that they are trying to come up with a MM system that works acceptably well both for parties and for solo players.
It’s fine in social modes. I just don’t want to see it in ranked play.

it should also be noted, as a to4 I don’t want to play 4 randoms. I’m not looking to stomp a team of randoms, I’d rather play a to4. That said, I believe teams of randoms, in ranked play, should be stomped. That’s what you get for searching a team playlist with no coms. It would be one thing if the randoms had coms and tried to work together but, when I search as a solo, that is not my experience. I am forever baffled as to why I am the only one talking. And I’m not a silver 4. I’m talking about random onyx level teammates with absolutely no mic OR they are in a party chat with someone else in an entirely different game. Playing at onyx level? Yeah ok. That guy deserves an advantage? Come on.
SGO SMACK wrote:
Can you please explain to me why you insist on catering to solo searching players in team playlists? The point of the team playlist is communicate and play like a team. If you can’t find 12, there is 8, if you can’t find 8 there is 4, if you can’t find 2, there is FFA. People who don’t want to communicate and don’t want to party up should expect to feel some pain. The people who are smart enough to play with a team in a team playlist should not be penalized because potatoes want to be potatoes. This applies to both arena opponents and wz teammates.
There have been solo players since Halo has been online so I don't know why people insist that solo players should only play in parties because it's never going to change.
RzR J3ST3R wrote:
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.
ZaedynFel threw up a good post to close the earlier thread on this where he noted the idea is to take the odds of full team winning against groups of solos from like 99:1 to 3:1. It's worth a read. Sounds like lots of game data had been looked at to determine where things need adjusting. I mean....99:1 isn't a game....it's shooting fish in a barrel.
This is the post I was commenting on to begin with. The 99:1 should motivate randoms to party up, or at the very least, plug their mics in. It shouldnt mean that we programmatically try solve what is perceived as a problem. This isn’t a problem, it’s exactly what should happen when 4 random dummies with no comms play against a coordinated team.

if all things are equal and the same pro team could play itself with one team having comms and they other having no mics, what do you think would happen?

this is insane that we are discussing or considering a change in the matchmaking algo for randoms. Half of the problem isn’t that they are randoms, it’s that they have no comms. Who’s fault is that? Part of this game is teamwork. Why would you give players who refuse to work as a team any advantage? This is literally insane. It’s a team playlist!
ZaedynFel wrote:
RzR J3ST3R wrote:
eLantern wrote:
SGO SMACK wrote:
If you assume a to4 diamond can play more like onyx’s and because of that, a to4 diamonds has to play against a to4 random onyx’s, they are going to get smoked.

this new system probably works decent for the silver vs gold level of players but a to4 low level onyx players will get stomped by a to4 random champs.
The new TrueSkill system (that isn't in place yet -- as far as I'm aware) that could account for adjustment tweaks toward a team's MMR value when the fireteam is a full party would be based on data which suggests the expected performance adjustment is warranted. How far that adjustment goes +/- is entirely dependent on the skill of the players in playing and communicating together verses what they're capable of when they're not playing together. Personally, I don't think the adjustment will ever be astronomical, but communication is a vital part of good high skill play so I think it could be more apparent at the upper middle levels and lower upper levels of play than anywhere else.
Yeah, @sgo_smack I don't think we're talking about a huge adjustment here. It might find that a To4 diamond 5 actually plays like diamond 6s when they party up. But I doubt it will mistake diamonds for champs.

As eLantern said, it's going to be data driven. If the game history data from the past year suggest that party-ing up gives you a 1-tier advantage, it will match you accordingly. In theory it won't go overboard.
ZaedynFel threw up a good post to close the earlier thread on this where he noted the idea is to take the odds of full team winning against groups of solos from like 99:1 to 3:1. It's worth a read. Sounds like lots of game data had been looked at to determine where things need adjusting. I mean....99:1 isn't a game....it's shooting fish in a barrel.
Right, like has been said, it's not an adhoc guess at how much better a team should be.

It's learned using a cutting edge statistical machine learning method that looks through 100s of millions of matches and learns, by playmode and party size, what that advantage is.

We've already evaluated the prototype on real matches and found it's very accurate at predicting the win probability. For example, it's around a half a tier for a lot of our Ranked 4v4 modes.

So we found, in our data, a mid-tier to4 Diamonds wins half their matches against low-Onyx solos.
Ugh. Does all of your data tell you who is communicating and who isn’t? Can you see who has their mic plugged in AND IS TALKING EFFECTIVELY ? I work with data for a living, I am a programmer. You can’t always rely on data alone and data can often be misinterpreted.
Again, just because a to4 diamonds can beat random onyx’s doesn’t mean you need balance that out. It means it’s working. You’re trying to solve a problem that doesn’t exist. You’re so hell bent on achieving close to a 50/50 win rate that your blind to the game and how it should work. Teams should beat randoms. Stop trying To “fix” that.
  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 4