So SBMM is designed to decide when you lose…?

[Darwi]

@Lenchy

Halo Tracker has a predictive algorithm similar to 343’s one to predict odds.

I googled it and found a post on the TRN forums;

@TRN Network Staff

it compares the average CSR of both teams

So my concerns are;

What algorithm? How do we know it’s similar to 343’s?

Why is it using CSR?

1. We know that after placement some people’s CSR can by 2-3 tiers below their MMR.

2. What about Smurfs (low CSR but rapidly rising MMR).

3. And “grinders” (CSR way over their MMR).

4. And weightings applied to squads? We know these are critical to balancing the teams.

5. And while MMR can rapidly respond to fluctuations in form - the CSR can’t. So you are going to lag a few games.

Overall, I honestly can’t see how it would be an accurate stat. If anyone has some better references for me to look at I would love to take a peek.

@TRN Network Staff

a difference of a full tier (for example Platinum 1 to Diamond 1) is pretty much a guaranteed win for the latter team.

The population of Halo 5 was constructed so that each standard deviation had a 25:75 chance vs the division above.

ie. A Diamond player would go 25:75 vs an Onyx player a division above.

Not quite “guaranteed”.

But anyway, I digress. I’m not putting a lot of faith in their algorithm if they rely on CSR ratings. Especially peri-placement.

It’s a fun stat for sure. But I’m not sure I would hang match making out to dry on the basis of it’s “results”.

[HEART_n_S0UL]

Does truskill2 count/deduct friendly fire?

[Lenchy]

What algorithm? How do we know it’s similar to 343’s?

Should be pretty standard. It compares number values (CSR), and does a ratio between both teams.

Why is it using CSR?

Guessing MMR not available.

1. We know that after placement some people’s CSR can by 2-3 tiers below their MMR.

After placements, sure. But, as described in 343’s blogpost about matcmaking, your CSR eventually arrives at a similar value to your MMR.

2. What about Smurfs (low CSR but rapidly rising MMR).

I think you’re vastly overestimating the number of smurfs in this game. This game doesn’t have the player base to make smurfing worth it.

3. And “grinders” (CSR way over their MMR).

These would be the remaining hardcore Halo fans. This is most likely the bigger part of the remaining 10k players. You also can’t really “grind” past your MMR in most multiplayer games. After a certain number of games, you settle with a close CSR/MMR.

5. And while MMR can rapidly respond to fluctuations in form - the CSR can’t. So you are going to lag a few games.

Your skill level also doesn’t fluctuate rapidly. It is usually a steady increase, until you stagnate.

Overall, I honestly can’t see how it would be an accurate stat. If anyone has some better references for me to look at I would love to take a peek.

Simple: the teams with the highest odds of winning, usually wins. The higher the rank, the more you contribute to increasing the odds.

The population of Halo 5 was constructed so that each standard deviation had a 25:75 chance vs the division above.

ie. A Diamond player would go 25:75 vs an Onyx player a division above.

Not quite “guaranteed”.

I’m not sure why you’re trying to debate every single point people are posting using niche examples: it is obvious from your above replies that you understand of CSR and MMR works, but it feels like you’re just choosing to ignore numbers and rely too heavily on outliers. Mentioning smurfs as an argument while there is likely only a few dozen players doing so. Yes, a Bronze player could beat a Diamond player, but odds are against that outcome.

While 343 can say that they aim for 50:50 matches, and while it is your opinion that the statement is factual, there is just so much data to support otherwise. 93:07 should never happen in ranked solo/duo. There is 0 way in hell that the teams couldn’t have been reshuffled to get the famous 50:50. There is tangible evidence, I’m not sure what more I could add.

[Darwi]

@Lenchy

Should be pretty standard.

What standard are we applying?

How can it even be close to 343s? - these likely average the team’s MMRs into a single curve and then subtract the two team curves to provide a new probability curve to reflect the result.

There is no way you can get that from the “average” CSR.

@Lenchy

It compares number values (CSR), and does a ratio between both teams.

And how do they determine the correct ratio apply?

I thought the key to HaloTracker not relying on the HaloDotAPI was that it had a very low memory footprint. To find the value to apply to the CSR delta would require a lot of game history.

@Lenchy

Guessing MMR not available.

Which is fine. It’s a stat site that’s primary function is to be fun and informative.

But it’s basing this stat on only a small part of the story (no MMR means, standard deviations, or weightings).

@Lenchy

But, as described in 343’s blogpost about matcmaking, your CSR eventually arrives at a similar value to your MMR.

It does - but people are getting particularly upset by it right now. A few days after a placing reset.

And while the CSR does indeed get closer - it can drift here and there.

And we still don’t have correct weightings (eg. for squads).

@Lenchy

I think you’re vastly overestimating the number of smurfs in this game. This game doesn’t have the player base to make smurfing worth it.

Not sure how the recent changes will affect them - but they were everywhere in the first half of season 2. Usually as part of the squad manipulations that helped Onyx players grind their CSR.

@Lenchy

You also can’t really “grind” past your MMR in most multiplayer games.

It’s been a significant problem in Halo Infinite. It’s one of the main reasons they recently introduced rank limits in squads.

@Lenchy

Your skill level also doesn’t fluctuate rapidly. It is usually a steady increase, until you stagnate.

Your form does fluctuate significantly. By up to as much as 150 points on any given day. And I’m wondering from some old waypoint pics generated by Josh Menke if Halo 5 / Infinite actually had some more game to game variability in MMR than we expected.

@Lenchy

Simple: the teams with the highest odds of winning, usually wins. The higher the rank, the more you contribute to increasing the odds.

But you still need to know the odds before you can increase them.

@Lenchy

but it feels like you’re just choosing to ignore numbers

No. Not ignoring any numbers.

I guess the basis of this discussion comes from why are we putting more validity on a 3rd party website making assumptions from CSR than the actual stat we have right in front of us - the average team MMR.

I might have to go over a few games and compare the % chance of winning from Halo Tracker with the MMR gap between each side. That might give some insight to how the % chance scales with MMR?

@Lenchy

there is just so much data to support otherwise. 93:07 should never happen in ranked solo/duo.

Speaking of outliers…

I would love to see what happened in this match! Both the teams involved and what the result was

Off hand I would say a squad has been waiting for too long and the match maker has had to go --yoink- it. This is the best we can do.

You can only hope that the 7% team salvaged something out of their afternoon.

@Lenchy

There is 0 way in hell that the teams couldn’t have been reshuffled to get the famous 50:50. There is tangible evidence, I’m not sure what more I could add.

I would imagine the lack of shuffling was more to do with squads than any deliberate attempt to sabotage one team.

Tangible indeed.

[Darwi]

And don’t get me wrong. I love HaloTracker. Use them a lot. They have saved my bacon in more Waypoint “discussions” than I can remember.

And they may, in fact, be validating their win % stat as they go. In which case - respect.

But all I’m saying is that at this point I’m not sure the stat is accurate enough that you can use it at absolute proof that 343 are manipulating games and trapping people at pre-designated ranks.

[eLantern]

@Darwi

@Lenchy

You also can’t really “grind” past your MMR in most multiplayer games.

It’s been a significant problem in Halo Infinite. It’s one of the main reasons they recently introduced rank limits in squads.

The CSR formula was changed by Josh Menke, and his team, some time back (during the somewhat latter stages of Halo 5) to specifically prevent runaway inflation/deflation of a person’s CSR. The change made it so that a 94% or greater win rate or loss rate would be required. For the vast majority of people this is completely impossible to achieve; particularly, where and when the population is healthy and especially after the implementation of TrueMatch as the matchmaker.

So as long as the 343 team didn’t reverse that change within Infinite’s CSR (and I don’t know why they would) it’d be extremely rare for anyone to grind their CSR well beyond the common range of their MMR fluctuations.

[Darwi]

@eLantern

So as long as the 343 team didn’t reverse that change within Infinite’s CSR (and I don’t know why they would) it’d be extremely rare for anyone to grind their CSR well beyond the common range of their MMR fluctuations.

They describe getting 0-1 points for the win and -15 for pretty much every loss (except when they get a team mate to quit out early).

So it’s definitely below their CSR.

But I’ll concede I don’t know how low. But we’ve seen graphs of MMR vs CSR showing they can drift apart up up to 75-90 points or so with natural play. So it’s conceivable that someone could manipulate and sustain something a bit higher?

[eLantern]

Yes, if a person is getting 0-1 CSR points for a win & -15 points for a loss then their MMR mean is below their current CSR by more than 15 points. And it will be unclear by exactly how much until they’re able to see a change in those CSR adjustment numbers.

Josh told me it wasn’t to unusual for MMR to fluctuate between +/- 150 points on the regular. This is because a person’s performance can range naturally from match-to-match due to countless variables.

So if a person has managed to converge their CSR to their MMR’s general mean, and the system has developed a strong certainty of that MMR mean (numerous performances consistently range within predictable limits around it), than it wouldn’t be all that uncommon to see smaller CSR adjustments in either direction.

If they’re seeing a bigger adjustment in one direction over another after an initial convergence it’s a sign that a person’s MMR mean has shifted due to a change in performance consistency and that the CSR is now attempting to re-converge. In other words, they’ve likely achieved a surprising win or loss along with a surprising individual performance; plus, repeated afterwards.

This is also how smurf accounts can get flagged. Purposefully playing subpar to establish a baseline MMR that’s below the player’s actual talent and then suddenly performing at their skill level will do several things. The system will adjust fairly quickly which prevents smurf accounts from being all that effective, but dramatic shifts will also flag the account for a potential ban.

The biggest annoyance is when this occurs in Ranked matches because that player may have earned a CSR that’s obviously significantly below their skill level and it may take considerable time for their CSR to converge to their quickly adjusted MMR. If that player is clueless regarding the systems in place (most probably are) they may think they’re continuing to fool the system due to their low CSR, but in actuality they’re just confusing the opposition and any websites offering predictive win percentages based on averaging team CSRs (such as HaloTracker apparently) which amounts to inaccurate guesswork, but still something potentially plausible.

And this is also why people need to be careful about sharing their character profile with others in their household. If there’s a distinct skill gap between individuals the profile could get flagged and banned.

[Dormiglione]

Lost a lot matches today. Got team mates who went all negative, while the whole other team went positive. Or i got put on server with high ping where nothing registered.
My conclusion is that a solo player should not play ranked. You need a team. But there are no after match lobbies, no spartan companies. I dont have anymore fun playing infinite.

[M1STA_WU1FY]

my conclusion is that everyone who wants to have fun should just delete the game

[ZETREN]

All of this means ABSOLUTELY NOTHING, when your shots, melees etc don’t register!!

The skilled are the people whose games register their inputs more consistently.

[M1STA_WU1FY]

I was about to say the same thing

This game’s ranking system is flawed as long as the game is so broken.

I have an anecdotal example. The last two games were against the same people, with the same teammates on my team. I went positive and got the most ball time the first game, but the second game

and I cannot emphasize this enough

My bullets weren’t working. Like, actually just plipping off people like I was throwing rocks at a train. No damage at all with probably 1/3 of the effort I put into killing people. It was the worst its been probably all week. Against the exact same people. I was shredding them like paper the game before, then suddenly I’m -10 the next match.

I saved about 5 clips of them just melting me through walls too, and my teammates, although they did better than me by a wide margin, were having the same problem.

As long as this game uses bad servers to influence the win-loss rates its not going to be fun to play.

[Crazyjacknuts]

You’re not crazy. I had a game earlier where the hit reg just took a vacation. Lol

[BurningWidow]

Lmao this happens so much.

[BansheemanX]

@Lenchy

67/33 is horrendous, considering the game could have balance teams to get closer to 50/50

Exactly. These are the games that the algorithm forces you to lose unless you’re having a miracle game where your weak link doesn’t throw.

It would make more sense if the odds were almost always 50-50. But you have to lose 50% of your games and if you’re good enough to solo influence wins, the algorithm will give you impossible odds.

The entire competitive halo community is mainly found on twitter. It doesn’t take long for you to find someone who is diamond - onyx 2000 posting a match where their odds are less than 10%

[Darwi]

@BansheemanX

These are the games that the algorithm forces you to lose unless you’re having a miracle game where your weak link doesn’t throw.

This is where hyperbole is letting your argument down.

1 in 3 is not horrendous.

It’s not ideal. I’m sure the system is working hard to keep it closer to 1 in 2. A more apt description is probably “frustrating”.

But it is not horrendous. It is far from impossible. You are not being forced to lose. A 1 in 3 chance does not require a miraculous performance to win. Particularly when you factor in all the variable that go together to make a team functional.

If only the people I play Poker against had the same mindset. I’d be rich.

@BansheemanX

the algorithm will give you impossible odds.

Hardly impossible.

And again. Do we really know how accurate those odds are. Basing them of the CSR doesn’t exactly instil me with confidence. Not saying they are completely inaccurate - but they are unlikely to be precise enough to prove anything.

What were the average team MMRs for those games?

And circling back to the original premise - how does some non 50-50 games prove that 343 is forcing you to not rank up? Apart from the anecdotal bias?

@BansheemanX

It doesn’t take long for you to find someone who is diamond - onyx 2000 posting a match where their odds are less than 10%

And?

I’d imagine the population matching for someone 2000+ would be nigh on impossible. Especially if and when they squad up with a buddy.

Is this any different for any other game out there? What do they specifically (and magically) do with their top 0.1% of players to make every single one of their matches crisp and even.

To demonstrate a deliberate strategy by 343 you are going to have to show that their matches are biased overall.

[Darwi]

As soon as I saw the conversation heading into comments about lag and desync I said to myself it won’t be long before someone adds that to the conspiracy theory;

@M1STA_WU1FY

As long as this game uses bad servers to influence the win-loss rates its not going to be fun to play.

And there we have it.

When did we all lose the ability to accept the fact that sometimes we just aren’t good enough.

[M1STA_WU1FY]

@Darwi

how does some non 50-50 games prove that 343 is forcing you to not rank up? Apart from the anecdotal bias?

its not “some”, its all of them. Every match is skewed in favor of one team. There are plenty of variables once the game is going, but when the algorithm predicts that red team has a 65% chance to win, its generally going to happen that way. The issue is that its not calculating that AFTER the match, its putting those odds in play on purpose by mixing the teams that specific way.

@Darwi

You are not being forced to lose.

You are definitely, with hard evidence to support it, being strongly encouraged to lose.

@Darwi

As soon as I saw the conversation heading into comments about lag and desync I said to myself it won’t be long before someone adds that to the conspiracy theory;

Its pretty easy to realize there’s a lot of patterns with how things go in this game. Your bullets will just stop working after a couple of good games. You’ll be shot through walls more often. You’ll have more whiffed rockets and shotgun shots.

Its because in addition to everything else, the game will encourage the team it wants to win to have the better server for the match as well. Or heck, maybe that’s partly how it calculates their win chance.

@Darwi

When did we all lose the ability to accept the fact that sometimes we just aren’t good enough.

Gonna need to explain this one, and choose your words very carefully

[Darwi]

@M1STA_WU1FY

its not “some”, its all of them.

It doesn’t matter if it’s all of them. But we know the match maker is struggling at the moment. And I really doubt that’s it’s actually “all” of them. Most of my placement matches this week have been plus or minus 20-30 MMR points.

It’s up to you prove your claim that the algorithm is deliberately assigning players one or the other. Which you haven’t so far.

@M1STA_WU1FY

when the algorithm predicts that red team has a 65% chance to win, its generally going to happen that way.

By generally, I assume you mean 2 out of every 3 times. Which is not ideal. But hardly an unscalable mountain.

And certainly better odds than no SBMM would give for average and below players to win.

@M1STA_WU1FY

The issue is that its not calculating that AFTER the match, its putting those odds in play on purpose by mixing the teams that specific way.

Again, it’s on you to prove these claims.

343 have categorically stated that the algorithms don’t.

I can’t see any pattern that suggests it’s anything but random.

You are calling 343 out as liars - so what’s your proof?

@M1STA_WU1FY

You are definitely, with hard evidence to support it, being strongly encouraged to lose.

So we are stepping back from forced to “strongly encouraged”.

Still a bit strong though, isn’t it?

And the “hard” evidence that you need to provide is that 343 are specifically picking who is going to be on the stronger team. It should be fairly easy to record your matches and use simple excel functions to find a trend. But you prefer to pick and choose games here and there as anecdotal proof.

@M1STA_WU1FY

Its pretty easy to realize there’s a lot of patterns with how things go in this game. Your bullets will just stop working after a couple of good games. You’ll be shot through walls more often. You’ll have more whiffed rockets and shotgun shots.

The game has a multitude of problems.

But there is only one pattern to recognise - and it’s random.

@M1STA_WU1FY

Gonna need to explain this one, and choose your words very carefully

[M1STA_WU1FY]

@Darwi

343 have categorically stated that the algorithms don’t.

I can’t see any pattern that suggests it’s anything but random.

They specifically make the matchmaker set up teams with the goal of a 50% win rate.

That means a 50% loss rate.

Checkmate

@Darwi

You are calling 343 out as liars - so what’s your proof?

You mean besides all the other times they’ve lied or been wrong about something?

@Darwi

Most of my placement matches this week have been plus or minus 20-30 MMR points.

How are you seeing your MMR?

@Darwi

But there is only one pattern to recognise - and it’s random.

It’s probably not, if it’s automated and utilizing advanced AI programs to establish matches. Unless you think the whole process it random then you honestly have no ability to refute the evidence.