So SBMM is designed to decide when you lose…?

Games Halo Infinite
447

The algorithm takes the players it has available and splits them into two even teams.

The 50:50 win / loss is a natural consequence of that. It’s the endpoint of a system that is ranking it’s player’s accurately.

The algorithm doesn’t know your current W/L. It doesn’t care if you have just won 10 in a row. Or lost 10 in a row. It just takes your current rank and splits the team’s as fairly as it can.

Your MMR could be 1200 going up, or 1200 going down. For the next game the only important bit is that it’s 1200.

So yes, it is doing it’s job.

What you are implying is that the algorithm takes into account your game history and then manipulates the result.

It doesn’t.

448

How does it accurately place you in a 50:50 matchup if it doesn’t know your game history?

How does your MMR not know your game history when your win rate and how many kills you have are how it’s calculated?

Do you even realize the contradiction?

449

“You will find that many of the truths we cling to depend greatly on our own point of view”

450

Your game history is essentially summarised in your MMR. With both it’s mean and width (standard deviation).

The MMR is more volatile than your CSR so it can go up and down on recent form (which reflects your recent W/L). And it’s width reflects how confident the system is in your rank.

And all these are taken into account when averaging the team’s MMRs, predicting the game result, and then allocating MMR change at the end of the game.

That’s why the MMR is a curve and not a single value.

The match maker doesn’t need to know the result of your last game. It just needs to know TrueSkill’s best guess at what your current rank is (mu) and how confident it is in that result (sigma).

Your MMR isn’t affected by your win rate per se.

You rank up by beating sides ranked above you and not losing to sides ranked below you.

And your KPM (and DPM) are weighted into the result.

You can actually rank up with a W/L less than 50% and rank down with a W/L greater than 50%. Especially early in the season when your curves are wide and volatile. But as your MMR matures that becomes harder to do.

Your global MMR has a detailed game history. That’s how it works. It shares this data to the offset MMRs for each playlist.

But the match maker doesn’t need to know all the back story. It just needs to know what your MMR currently is. Your current MMR reflects your current form better than your CSR - which is designed to move slower and smooth out your ranking journey.

2 Likes
451

So to summarize, the algorithm must be very aware of your past performance and therefore react accordingly to balance out the 50% win/loss goal.

All right there in black and white

452

You keep looking at the end point and using it to justify the means.

We have the same end point; 50:50 chance of winning.

But very different means; balanced teams vs manipulated teams.

It is true that your recent performance is reflected in your MMR and since the algorithm uses your MMR to match you can say that recent performance influences matching.

But the key point is that the match maker has no idea what past performance led to your current MMR. It doesn’t know, or care, if it’s on the back of 10 wins or 10 losses. Or a run or W-L-W-L …

It takes your MMR and balances the next game the best it can.

It doesn’t know if you won or lost your last game. And it doesn’t need to.

The match maker knows your current form. And that’s good. That’s the only information it needs. Whether that’s good form or bad form in respect to the player is irrelevant. It just needs to know how you are going right now so it can create the best match for everyone in that game.

2 Likes
453

The teams aren’t balanced though, someone always has a weighted chance to win. That’s why you will randomly get low ranked people on your team while you’re on a win streak, in order to bring you down.

454

They are as balanced as they can be.

Given squads etc.

The danger is applying this chance retrospectively. You can’t assume the match maker knew something was going to happen.

Someone performing better or worse than predicted.

In fact everyone probably performs a bit better or worse. And the result leans on the sum total of all those variances.

On those (hopefully rare) occasions where there are squads or not a fair way of dividing the players then this mismatch is taken into account with the result vs your CSR and MMR.

Key here is random.

There is no allocations on the basis of streaks.

The match maker doesn’t know and doesn’t care.

It may feel personal, especially after a win or two. But it’s not.

455

Stupid question.

Why is 50% the goal? Is there any evidence that it’s more fun?

like… as longs as the sbmm exists and is designed to mostly put players into games where they have an overall chance of winning about as many games as they lose… why shoot for 50 50 each match instead of going something like 45 50 55 over the course of three games?

456

But it does know the chance beforehand. Why else would that stat exist?

That doesn’t make sense if the matchmaker wasn’t making the match lopsided to begin with. It would only know to give a buffer if it knew it had a match that wasn’t evenly set up.

Yes it does. That’s it sole reason for existing. It’s not like it’s going to half -Yoink!- it.

Studies show that this is the best way to increase player retention.

The problem is that the retention is due to people being pissed off and playing more to make up the difference from lost progress and not due to people actually enjoying the game. It’s basically like a gambling addiction.

“Sooner or later I have to win, I’ve lost too much to quit now”

457

Correct. The ranking system is designed for a healthy population. Not a game that’s already under 10K players.

Matches will continue to be random. good players will have to carry more and end up losing more.

probably not. More likely that there’s nobody else to match with in the Queue on the server you will be playing on.
It’ll be random matching until tens of thousands of players start playing regularly again.

458

also, it occurs to me that without some sort of mode like firefight- a lack of multiplayer pve- this issue is made worse than it needs to be, and retention is worsened. The concept of balancing social games so you win some and you lose some isn’t an inherently bad idea, but it does seem to result in tighter matches- and some people don’t want constantly tight matches, they want fun stuff. Social mode lacks social options, and people who don’t want tight matches don’t really have a way to use infinite right now except by using campaign, sorta.

1 Like
459

The evidence is that people are more likely to quit out of a session (for more than 8 hours) after a lopsided game.

More than poor performance, losing streaks, wait to match, or pings.

Plus the evidence they have for player retention overall. It’s not the year 2000 - people don’t want to be thrashed and humiliated into “getting gud”.

As for 50:50 vs other ratios - I don’t know. At what point does a player start to feel they didn’t have a chance?

Maybe 40:60 is where it starts to unravel?

There are so many more variables to what makes a team than just MMR matching - so all the other factors tend to smudge the game about anyway. While it tries to make it 50:50 the reality is that no game ever is.

The match maker knows the stat; eg. 2% advantage to Blue team. But that’s pre game. It doesn’t know that Player 3 on Red team is going to have a stinker and drag his team down.

This only happens when the players available can’t be divided evenly. Like squads or the last eight players in the queue waiting for a game.

You can’t divide a player in half and put a bit on each team to even it out.

It doesn’t need to know the result of your last game.

Why would it?

It just needs to know your rank.

It jars some people’s experience.

Most people are able to go out there and have fun and still revel in the contest. They don’t blow a gasket when the team goes down. They can swap up the intensity of ranked vs social.

Some people want to go at 100% but not have to sweat. ie. play opponents lower than them. The problem is that isn’t fun for their opponents.

More options to let off steam would be welcome. eg. multi-team, PVE, etc.

1 Like
460

Some people want to go at 100% but not have to sweat. ie. play opponents lower than them. The problem is that isn’t fun for their opponents.

I don’t really care about these people. I don’t mean that in a mean way, I just mean that what they want is what the rest of the community doesn’t. This is a social game- if it doesn’t adapt to social forces, it dies.

Most people are able to go out there and have fun and still revel in the contest. They don’t blow a gasket when the team goes down. They can swap up the intensity of ranked vs social.

how to phrase this… workday hard, long, wanna relax and have fun. Tight matches are fun that require tight focus. It isn’t just about wanting to win, but about wanting to not let your team down. Nobody wants to be that guy that causes the team to fail.

Also, I think i just found a weakness in the model. I would bet money there are certain weekdays (probably tuesday and fridays) where the model can’t figure out people reliably, because people who normally have good scores are chasing weeklies, and matches that should be wins become loses because the designated carry player is going after the weekly, and vice versa.

Last I’ll say about it, but I’m also figuring there’s more reasons for which sbmm isn’t to blame, nor is the 50% target. You’re winning 50% of matches on average, right? But how does it feel to win them? It feels… okay. And how does it feel to lose them? It feels disappointing.

I don’t think halo necessarily needs more ‘turnaround’ points in the gameplay, but i’m starting to wonder if the reason that the games feel so tight is because the games are just less fun than in older games. Infection was huge fun. Griffball was fun. Slayer was fun, but not over and over and over. you get people who play the oddball matches because they have to for the weekly, and then they just stand there.

sbmm by itself is fine. But i think the weekly challenge as a concept is killing the fun by virtue of putting you with players who ordinarily would be good at the game mode or bad at the game mode as appropriate, but those players are only there for the weekly and aren’t going to engage in a mode they don’t care about playing because it bores them.

461

Your rank changes based on the outcome of your last game

It does, or it wouldn’t place them on a team specifically so they would lose

462

Same page.

Keep in mind we’re not talking every game going 50-49 in the dying seconds.

They just have to be balanced to the point that both teams felt they had a chance.

Unless you are really playing off key it’s unlikely anyone would notice.

And we’re talking in respect to your Social MMR here. Not your Ranked MMR.

Yep. A lot of the challenges are inherently bad. Or at least the way people play them is bad.

Don’t have an easy solution for that life problem.

Going in with a different expectation is a good start.

We certainly need more low key games. Like Infection.

Or my favourite; multi-team. It’s hard to play ranked type sweaty when you are out numbered 2:1. And there are two losers each match so it’s less stressful.

Weekly challenge needs to be score based.

With multipliers to encourage game types or events.

But that way someone who really doesn’t want to be in Fiesta can keep the score ticking in whatever mode better floats their boat.

463

It does.

But the match maker doesn’t need to know the actual outcome that got you there.

It doesn’t care if you just hit 1200 on the back of a come from behind win - or you have just plummeted to 1200 on the back of a devastating loss.

It only cares that you are currently operating as a 1200. That’s what it needs for this game. That’s the pidgeon hole you have been allotted to.

It may place them on a team that has a slightly reduced chance to win.

But it doesn’t know how each player is going to hold up.

And we are talking minor advantages here. We’re not using HaloTracker’s dubious stat (which we have shown didn’t hold up in over half the games we sampled).

464

Low population broke the SBMM.
Still no word on what the SBMM experiment entails. Just that it lasts until the end of the season.

465

Was it a SBMM experiment?

I thought it was more along the lines of a CSR behaviour modification.

I doubt they would do much to TrueSkill2 / MMR.

We may never know…

466

In reality it probably works out this way. For a variety of reasons.

I’d expect any competitive player would view it as acceptable. When they’re favored they go in expecting to win. When the odds are even they go in expecting to win. When they’re unfavored they’re expecting to win. If it doesn’t work out this way in any of these circumstances they get knocked down, get up, brush the dust off their big boy pants and look toward improvement.

Yet, players are supposedly quitting in droves because they feel the MM is rigged to achieve a 50/50 win/loss. One would think there are some contradictions hiding in the bushes if they were to look closely enough.

At this point I’d have to ask a genuine question. What exactly are you expecting to achieve with this crusade declaring the MM rigged? Take a moment to seriously think about it. If it is rigged the likely response is to slap working as intended on it. If it isn’t there is likely to be some eye rolling involved. Either way, no progress.

If the thought is too many games are lopsided, that’s a valid complaint. If the thought is players feel pressured to carry teammates into the stratosphere too often, valid complaint. Declaring it rigged doesn’t need a place in those discussions. Just make those criticisms and toss the ball into 343’s court. Yeah, they’re probably picking it up and running to the wrong endzone. The point still stands.

2 Likes