This ranking system has to change

Games Halo Infinite
21

I agree.

But what TrueSkill does is represent your skill as a curve (it never pretends to know your exact rank). It just assumes your rank is somewhere on that curve (most likely around the mean).

So you can match around the overlapping of the curves.

Someone who is Platinum 5 with a very narrow curve - the system knows they are around Platinum 5 and can try and match you with other Platinum 5’s. But someone who has a broad curve, from Plat 2 to Diamond 2, could very well be a Platinum 5 - so the system happily matches them while it gains more confidence in the rank.

The problems is that the second player will have a low CSR (which can upset people off the bat), and while statistically unlikely - they could be closer to Gold 6 or Diamond 2 than the system wants, and some uneven ranking comes into the match.

This is part of the reason why players in placing are “unranked”.

But the upside is that players can be quickly and efficiently matched. Much faster than a simple win/loss system (which could take 100’s of games to sort players).

But later in the season, when everyone has a mature rank, match making should be a lot more consistent. But that is also the time that people start complaining that they aren’t ranking up!

After placement some players will still have quite wide curves… so their CSR may not represent their true rank. Although this should really only affect new players as experienced players go into placement with their MMR and if they play consistently should come out with a fairly narrow MMR curve similar to the one they went in with.

And we are also heavily beholden to match-making. You can only match players who are available at the time.

Plus they are still allowing squads to go in with wide ranges of skills. Personally I wish they wouldn’t (my recent placement matches were terrible because of this). That also spreads the ranks across the board.

22

I think they wanted to make a ranking system so sophisticated that in the end they messed up the algorithm and got completely incoherent results.
In other words, the idea was good, but put into practice it is a disaster. You no longer know why you go up or why you go down. I don’t know, maybe something so dense wasn’t needed. I no longer play ranked on infinite.

23

the only thing skill is a ladder not a curve

24

This is a big problem.

If they are going to show the change in number (and I wish they hadn’t) - it should be fully transparent.

Simple W/L sorting takes 100’s of games to sort out the final rank. It would be a disaster in regards to the quality of match-making.

ELO systems may be “dense” but they can find your rank in less than 50 games (and ball park you in less than 10). They are efficient and accurate in what they do - so you may as well use them.

But that doesn’t necessarily translate into a good CSR system.

25

I understand you and I understand the theory but look, for example, I have a game in which I win, I am in first place. 10 kills - 3 deaths, K/D Ratio 3.33… Result: Downgraded. Well, I can’t understand these things. But it’s not just this, there are hundreds of inconsistencies in rankings and matchups.

1 Like
26

This wouldn’t happen.

Basically if you win you rank up and if you lose you rank down.

How much your rank changes depends on;

  1. The relative skill of the opposing team. Expected results don’t force much change, while upsets will tend to push your rank.

  2. The shape of your MMR curve. Immature curves (new or around placing) are wide and volatile - they want to move. While mature curves are tall, narrow, and static - they tend to resist change.

  3. Metrics and weightings. eg. kills/minute. We don’t know for sure how these are applied across different game types.

  4. CSR reconciliation. The nature of any W/L system is that it will oscillate. You will drift up and down with form. But Infinite has a positive bias - rewarding every win. If you have a run of expected wins your CSR will tend to drift ahead of your MMR and the next loss may cause a heavier than expected fall.

So, in regards to your theoretical example;

You won’t be downgraded. You won.

Yay. And this may push you up more than if you were in a lower placing. But if the win was expected your MMR isn’t going anywhere.

Would be lovely. But your individual performance is trumped by the result (win or loss) and the expectation of the result. Doing (relatively) badly in an upset win is likely to be better for rank then doing well in an expected thrashing.

And keep in mind that the metric is probably K/min rather than K/D. Going 5:4 in a two minute game (2.5 K/min) is better than going 10:3 in a ten minute game (1.0 K/min). It’s a subtle difference.

But when you think about it K/min is better at identifying the big fish in a small pond - so is a useful metric for ranking you faster (as opposed to higher). As you match more equivalent opponents your K/min will fall much harder than your K/D.

There aren’t as many as you would think.

There is a lot we don’t know - metrics, weightings, applicability across game types, etc - but generally most behaviours can be explained if you look at your game history.

A lot of people still get confused over the ranking bar and the fact that it shows per session and not just what happened in the last game.

It’s frustrating when you think someone else is being rewarded more than you for the same win - but it all comes down to the shape of their curve (they are on a different part of the MMR journey), the metrics applied (which we aren’t fully privy to), and the push/pull of your MMR on your CSR.

And what I really don’t get is this angst over a handful of CSR points. We would be much better served if they stayed hidden and we just enjoyed being Bronze 1 to Onyx 6.

The match-making algorithms are a different beast entirely.

In a perfect world every game would be buttery smooth in terms of match-making - but the system is beholden to who is available at the time. And it can’t predict who is going to play well with whom. And then you throw in squads with wide skill ranges and well… it’s a logistical minefield. And that’s before you even start on Smurfs and squad manipulations.

27

Okay Darwi. If you need so much development to explain how it works to me, it’s definitely overly sophisticated.
What doesn’t seem right to me is that you say “that can’t happen”. You have at your disposal my statistics in Halotracker and you can see it. You can also see my service record in halowaypoint. I have no need to lie.
Also tell you that I am not a new player in Halo. I’ve been playing this game for 14 years and I’ve played them all.
From Europe it has always been difficult to play multiplayer and ranked when it was P2P and it got worse with dedicated servers in Halo 5.
I play a couple of hours a day and some weekends so I don’t pretend to be a professional or reach onyx and things like that . I’m just trying to have fun.
But without a doubt, Halo Infinite’s ranking system doesn’t work well, it’s inconsistent, it’s not very understandable for players. And not only if we compare it with previous Halos but with almost any current game. But it’s not that it’s a particular opinion of mine, it’s something “vox populi”.
Still, I appreciate the explanation but after playing almost 1700 games of this Halo Infinite I think I have enough experience to say that the ranking system, the matchmaking, the servers, the lag and the desynchronism, makes this Halo a real disaster. Thanks for all and sorry for the translation.

29

I guess it is sophisticated. The calculus type maths bewilders me. :slight_smile:

But essentially it’s just a clever algorithm to work out how much rank to add or take to find everyone’s rank as quickly as possible.

It can do in a few games what would otherwise take 100’s.

I apologise profusely. I didn’t realise you were putting up a real world example.

But looking at Halo Tracker it just says unranked -1 for that game. I think the -1 is a glitch from Halo Tracker when the API returns a non-sensical value for the rank. On the Halo 5 version it used to return “null” in the xml. On the server side it may be a change in the rounding that returns a -1 as opposed to a 0.

Bottom line is I don’t think you actually lost rank in that game.

With you all the way.

It works well and is not as inconsistent as people make out.

But I agree they have done a -yoink- poor job at explaining it. Or even selling the silly way they use a large scale number on the CSR. To a large extent they are shooting themselves in the foot.

At least in H5 we had someone from 343 on the forums answering questions, putting up FAQ threads, and even looking up peoples MMR for them.

I’m happy to stand up for the ranking system. TrueSkill is a thing of beauty.

But I agree that the CSR needs reworking.

And I can’t defend the match-making, servers, lag, and desync. Luckily I don’t seem to be a huge victim of most of those - but I know they are creating havoc for a lot of people.