That rank system blog post wasn't very good or helpful

[Darwi]

@TheArlstocrats

The MASSIVE problem with that is, people don’t play the same with social playlists as they do ranked. For example, I play with friends (sometimes having drinks) when I hit up social and BTB. I play ranked very differently, as do most people. Can someone explain why my performance in a fun, casual mode should have any bearing whatsoever on my rewards in a ranked system?

I don’t think it does.

Each playlist has it’s own offset data - which should function as it’s own MMR.

We know that Halo 5, which had more than one ranked playlist, had clearly separate MMRs.

Anecdotally I’ve had no change in ranked after playing bots or mucking around in BTB with mates (we are bad).

We know that a seeding MMR is used when entering a new playlist. This is different from going back and forth on a mature account.

@TheArlstocrats

The solution is simple - CSR vs CSR. Stop trying to outsmart your players.

Seeing as your CSR reflects your MMR this is what you are pretty much doing.

They are only significantly different post placement. After that they stick to each other pretty closely (your MMR pulls on your CSR).

The solution is even simpler. Keep it MMR vs MMR (after all, this is the more accurate measurement).

But given the MMR is a curve (not an acutal number) you still need a CSR to describe it. eg. 1 to 50.

@Primus_Ego_Sum

Their system, which matches everything based on a global hidden MMR that everything in the game affects, is absolute nonsensical trash and they refuse to listen to that kind of feedback.

Still haven’t seen any clear proof of this globally acting MMR.

@Dead_End_24

The blog wasn’t bad but I’m VERY disappointed and not happy at all to hear the social play affects your MMR.

Again, what they described is borrowing your social MMR when you go into ranked for the first time. This is a feature of TrueSkill2 that makes sense when you have multiple ranked playlists. If you are Diamond 3 in Ranked Slayer and you go into Ranked FFA for the first time, it makes sense to use Diamond 3 as a starting point.

Because Infinite only has Social, Bots, and Ranked - it back fires a bit.

But again. It’s only new accounts… and rankings sorts things out pretty quickly.

When the Reddit player played Bots and then went into a new Ranked account they came up against 2000+ Onyx players - but their rank fell very quickly as TrueSkill2 reconciled their performance against expectation.

But when a follow up player using a mature account played Bots for a few hours to get an insanely high KDA they found no change when they went back into Ranked.

@Dead_End_24

A other thing I “loved” (not really) is how they didn’t even say everything that affects trueskill2. They said "TrueSkill2 uses a variety of data about each player’s performance in multiplayer matches (including wins, losses, kills, deaths, etc.)

I would love a detailed list of the weightings and how they are applied… but I guess they are being careful.

We are already seeing how toxic players are getting over their K/D (even though it’s probably not helping that much).

Hopefully they can give us a bit more info in another post. I would love to know how they look at objective games.

[xLEFx777x]

I actually fell asleep reading it. True story.

[BlueDevistator]

Dear 343, if I can play at least a dozen matches of swat, and lose every single one while usually being the best performing player in the match, then your MMR system could use some improvement.

[B84_W56_H88]

Feels in line with every other blogpost from them. I’m positive that 343 didn’t develop the game and are just throwing around the IP and see what sticks.

[SepticonWarrior]

@Darwi

@SepticonWarrior

I don’t think 10 games can determine fair matches,

It can.

Sure, it needs closer to 50 to be super sure. But it can ball park most people in a handful of games.

I don’t know how long you’ve played ranked online matchmaking, but even in Halo 2’s fan-fav ranking system, if you were a decent player in the mid-30 ranks (which was typical back then), you likely played several games where you could dominate as well as several games where it was unbearable (due to many aspects) where you get crushed. - attempting to perfect a system so restrictive, as they attempt with Infinite, has been known to wreck entire civilizations!

@Darwi

@SepticonWarrior

or B) matched vs Onyx players

Only going to happen if you’re playing at Onyx level.

Did they fix something? Honest question, cause my first time through unranked was a painful experience where I could barely get kills. Then, long and behold, after that gruesome stage, it was better where I felt that I was contributing. Some would say it was better than social, but BR starts play a role in that as well.

@Darwi

@SepticonWarrior

Ranked needs players to have that period in their ranks where they feel like they’re worth a damn in the game, so they become slightly addicted to performing

Ranked needs to rank you as quickly and as efficiently as it can… so it can get on with it’s job of match making.

Which group of players do you suggest they use to boost your ego while you wait? How will this help them to become addicted to the game?

Ranked doesn’t necessarily need to rank you quickly. What usually happens when you attempt this is a system where you can win several games in a row to give you the feeling of ranking up, then lose one, and it think you should be knocked right back down where you started - because it thinks you should be at that particular spot on the ranks.

They shouldn’t use a group of new players so you can stomp them, although that would be ideal anyway.
History has shown (ie LAN is better than Matchmaking) that mixing teams of various skill levels makes for the best experience for everyone. CE was so good in college because we would have teams with 1 good player, 2 average, and 1 noob on each team, which made for some hilarious moments. I still think that’s the best formula, for at least social playlists… but I can see why they want to attempt on placing all equal skills in the match as well, not that I agree with that approach.

[Darwi]

@BlueDevistator

if I can play at least a dozen matches of swat, and lose every single one while usually being the best performing player in the match, then your MMR system could use some improvement.

Halo Tracker has your SWAT record at 54% W/L and a K/D of just over 1.

Matchmaking seems to be doing ok?

[Darwi]

@SepticonWarrior

I don’t know how long you’ve played ranked online matchmaking, but even in Halo 2’s fan-fav ranking system

I’ve played Halo since day dot… But didn’t really start online until Halo 3.

I’ve seen a lot of people recall H2 fondly… Although a few also acknowledged how slow it was to rank up… And how uneven the matches could be.

I do like the way the last 10 levels were named and had symbols. It could be cool if they did this for Onyx.

@SepticonWarrior

Did they fix something? Honest question, cause my first time through unranked was a painful experience where I could barely get kills

The first ranking run through is always going to be a bit hit and miss. Everyone is unranked. I would think you’ll find subsequent placements a littler smoother.

@SepticonWarrior

What usually happens when you attempt this is a system where you can win several games in a row to give you the feeling of ranking up, then lose one, and it think you should be knocked right back down where you started - because it thinks you should be at that particular spot on the ranks.

This oscillation pattern of a few steps forward followed by a step back is the nature of CSR system with a positive bias. The system feels compelled to give you CSR for a win, even if the team you beat was rubbish.

Your CSR gets ahead of your MMR - which waits patiently for your next loss to drag you back.

And it’s further compounded of you are a good player or running in a good squad. The match maker may struggle to find better teams to put you against. You’ll have more easy / even games than hard ones.

@SepticonWarrior

History has shown (ie LAN is better than Matchmaking) that mixing teams of various skill levels makes for the best experience for everyone.

Yep. We used to have Halo LAN parties after work.

But the key here is that you are usually dividing the teams with an intricate knowledge of the players involved. You have way more info than the match maker does.

Same now for weekly customs get together. We’re mixing Onyx to Silver… but we can make it work because we know which combos work best.

[Dead_End_24]

[quote=“Dead_End_24, post:19, topic:508303”]
The blog wasn’t bad but I’m VERY disappointed and not happy at all to hear the social play affects your MMR.[/quote]

Again, what they described is borrowing your social MMR when you go into ranked for the first time. This is a feature of TrueSkill2 that makes sense when you have multiple ranked playlists. If you are Diamond 3 in Ranked Slayer and you go into Ranked FFA for the first time, it makes sense to use Diamond 3 as a starting point.

Because Infinite only has Social, Bots, and Ranked - it back fires a bit.

But again. It’s only new accounts… and rankings sorts things out pretty quickly.

When the Reddit player played Bots and then went into a new Ranked account they came up against 2000+ Onyx players - but their rank fell very quickly as TrueSkill2 reconciled their performance against expectation.

But when a follow up player using a mature account played Bots for a few hours to get an insanely high KDA they found no change when they went back into Ranked.

And I would be ok if they wanted to take my say rank from a other ranked playlist, say team Slayer and use it in ranked FFA, but they aren’t saying that necessarily. They’re saying if you go into a ranked playlist for the first time they will take your social stats to use as a base. Seeing as how there’s only one ranked playlist right now, that’s what they’re doing. If in the future they have multiple ranked playlists and they take it from another ranked playlist only, then I’m perfectly fine with that. As of now, that is in the case and in my opinion I don’t think that’s right at all.

@Dead_End_24

A other thing I “loved” (not really) is how they didn’t even say everything that affects trueskill2. They said "TrueSkill2 uses a variety of data about each player’s performance in multiplayer matches (including wins, losses, kills, deaths, etc.)

I would love a detailed list of the weightings and how they are applied… but I guess they are being careful.

We are already seeing how toxic players are getting over their K/D (even though it’s probably not helping that much).

Hopefully they can give us a bit more info in another post. I would love to know how they look at objective games.
[/quote]

It’s ridiculous they haven’t told us exactly what weighs on it. Probably because back when Halo 5 when it was found out that things like assist don’t matter at all people were pissed off about it and played differently. They are most likely afraid to release exactly what matters and what doesn’t as they know people will play differently.

I know this for a fact, as I was in many many many discussions with Josh Menke and the community about how ranked should be for years. Myself and many others were not happy with the fact that’s something like a assist didn’t matter and I gave many examples, which Josh Meeks fully acknowledged, that assist were just as important as say kills.

I can’t stand this whole ranking system based on individual performance when it’s a team playlist. I’ve given countless scenarios and situations as to why it’s stupid to base your rank on individual play in a team playlist. I’ve already expressed my views on this for years and even gave Josh my solution on the ranking system and what to do and he didn’t disagree that it was a bad idea and that they should look into it… without getting into it heavily it was basically you have two ranks. You have your overall rank in ranked which is based on your individual performance (Pretty much like what they do now with some changes) and then you have your rank that is purely based on wins and losses for each ranked playlist. So for example you could be a diamond 2 and ranked slayer (Base purely on wins and losses like the old Halo games) but your individual rank is maybe say diamond 5. Meaning you’re a good player but maybe you don’t play very well as a team hence your win-loss record is not very good…but that matters because, well he left 343 so…yeah.

I’m getting to the point where I’m probably not going to play ranked anywhere near as much in this game (which if you knew me, you would know that that’s my thing) and probably just the game in general. This game just has so many issues. It’s sad Sucks too, because I do like the base gameplay and I want to like the game a lot just like many other people but It’s really hard to right now. Maybe in a year or two it will be a lot better but as of now it’s been pretty disappointing.

PS, I hate how you quotes stuff in this, Sorry if it’s confusing. I tried to make it less confusing with bold

[Primus_Ego_Sum]

The reality here is simple.

1. If your ranking system has a global hidden MMR that is affected by all playlists, your ranking system is bad.

2. If your ranking system doesn’t use visible rank to match players in the ranked list(s) - thus making that visible rank meaningless - your ranking system is bad.

3. If your ranking system uses a similar strictness of SBMM in social and ranked, your ranking system is bad.

I would say a ranking system is bad if it does only one of those things. Infinite’s does all three. It’s the worst ranked/SBMM implementation in the industry.

Infinite needs per-playlist mmr, it needs to match players based on CSR in the ranked list, and it needs to stop treating game modes like quick play, Fiesta, and swat as if they are competitive ranked modes and prioritize connection over SBMM.

[benjiFatStax]

Do you have a sense for whether W/L affects MMR, or just CSR? I think my “aha” moment with this update was that:

a) CSR truly is truly irrelevant, it’s a purely aesthetic badge that may or may not be close to your MMR at any given point in time and is not used in matchmaking algorithms. Despite the assertion that it should closely match after 50 games, I’ve personally experienced substantial drift which Ken and I worked through in that other thread–drift that appeared to be over 100 CSR/MMR points despite being hundreds of games in.

b) Based on that, and the fact that MMR seems more oriented towards individual performance (as we saw, it can go up in losses and down in wins), and the only individual performance we can seem to observe ourselves (and of course it’s suggested in the paper) is K/Min, then…

c) The “sweat” strategy is the correct way to play ranked. You should slay above all else, as that appears to have the most effect on your “real” rank. It is much more beneficial in the long run to slay and lose than to play conservative and win. Especially relevant in Slayer, where playing fast, going 13/15, and losing would be preferable from an MMR perspective to playing slow for 8/5 and winning.

That make sense? Or am I off base once again?

[BlueDevistator]

Works ok most of the time, but definitely felt broken last night. A very noticeable change at midnight.
It was probably just an unlucky streak, but angered me nonetheless.

[Little_Squash]

@Primus_Ego_Sum

Infinite needs per-playlist mmr, it needs to match players based on CSR in the ranked list, and it needs to stop treating game modes like quick play, Fiesta, and swat as if they are competitive ranked modes and prioritize connection over SBMM.

But since we explained how it works, everyone loves it! Plus it helps retain players! And if you don’t love it, just trust us, you really do and you just don’t know it yet.

-343 (probably)

[II_R3GR3T_II]

What you mean the fact that once your in “Onyx” it’s like impossible to lose your rank! I have seen Onyx players lose like 10 games in a row they do absolutely horrible and lose 1-2 csr… Plat and Diamond you lose 10 in a row your will loss over 150 csr… something is completely broken…

[Darwi]

@Dead_End_24

And I would be ok if they wanted to take my say rank from a other ranked playlist, say team Slayer and use it in ranked FFA, but they aren’t saying that necessarily. They’re saying if you go into a ranked playlist for the first time they will take your social stats to use as a base. Seeing as how there’s only one ranked playlist right now, that’s what they’re doing. If in the future they have multiple ranked playlists and they take it from another ranked playlist only, then I’m perfectly fine with that. As of now, that is in the case and in my opinion I don’t think that’s right at all.

The paper on TrueSkill2 mentions using the most similar playlist available. So I imagine they will set it to ranked ones in the future - like they would have in H5.

The bots to ranked anomaly is pretty funny when you think about it. It’s harmless. Placement will sort your true rank out pretty quickly.

@Dead_End_24

It’s ridiculous they haven’t told us exactly what weighs on it

I can understand why they don’t.

@Dead_End_24

Myself and many others were not happy with the fact that’s something like a assist didn’t matter and I gave many examples

Yep, been a bugbear of mine too.

Hopefully we can do something with damage ratios in the future. For and against. Effective (ended in a kill vs escaped). Early vs finishing.

@Dead_End_24

PS, I hate how you quotes stuff in this, Sorry if it’s confusing. I tried to make it less confusing with bold

The key is to have the quote and end quote tags on a separate line.

[Farsight_360]

@Darwi have you seen Max Hoberman’s responses to Mint Blitz on twitter? I know we’ve gone back and forth here (agreeing in some respects and disagreeing on others) – but wanted to flag it in particular because they are talking about Trueskills strengths and weaknesses. Good stuff overall. You will probably have to go to Hoberman’s tweets and replies to see everything.

https://twitter.com/maxhoberman/status/1502633023497555974?s=21

[Primus_Ego_Sum]

@Little_Squash

But since we explained how it works, everyone loves it! Plus it helps retain players! And if you don’t love it, just trust us, you really do and you just don’t know it yet.

I love how 343’s response to two of this game’s biggest problems (garbage netcode, garbage SBMM/rank systems) has just been to give a basic explanation of their decisions. They just ignore the fact that their particular decisions are in fact the problem. They are experts at gaslighting their fanbase.

This game has the worst netcode in the industry. What does 343 do? Write a blog post explaining internet latency 101.

This game has an unnecessary, confusing, and frustrating SBMM implementation that nobody except the devs and Darwi like and is worse than just about any other system out there. What does 343 do? Write a blog post explaining their basic decision making process for how and why they developed a terrible system.

[Zelliam]

And yet Staten comes out last week saying such people are having…
“… life and work balance issues…”

Yet they smother these forums rebuking critique that could or should be seen as positive feedback.
After all, the motivation behind every negative post, or dis-pleased fan taking the time to voice it here. Other than just moving onto another game- is due to a love for the product that is halo, and the hope it can improve.

[CSevenZer0]

Josh Monke - the Halo SBMM and Ranking systems are the best ever developed. If you don’t like them you’re just a bad player. The data tells me so.

Player - do you even play the game though?

Josh Monke - get gud

[DDoubleDDeez479]

The blog was pretty bland; just standard info and a bunch os stuff that is already common knowledge.

@jxsh003

I really don’t understand a system where hidden stats are what dictate ranked matches, not the RANKS themselves

Yeah, this is the big issue. It leads to the system feeling especially miserable as doing well in a lost game boosts a number the player doesn’t see, which then works to ensure subsequent games get harder and reward less csr. It also makes the system easy to manipulate to get easy games and boost. If the mmr was just used to rate performance and adjust the csr gain/loss, there would not be a problem.

343 seems to be treating ranked as though the objective is to get perfectly balanced games right from the start. It isn’t.

The purpose is to create a competitive environment where people do their best to win and are rewarded with rank gains for being or becoming better at the game. Trying to artificially “balance” every game with hidden numbers just makes it harder to climb or drop to where you belong, while games end up wildly unbalanced because they are using the less stable performance tracker number to put together matches.

No radical overhaul needed, just another instance of 343 putting their priorities in reverse order. All they have to do is stop trying to reinvent the wheel by turning it on its side and everything will work just fine.

[DDoubleDDeez479]

@Primus_Ego_Sum

The reality here is simple.

1. If your ranking system has a global hidden MMR that is affected by all playlists, your ranking system is bad.
2. If your ranking system doesn’t use visible rank to match players in the ranked list(s) - thus making that visible rank meaningless - your ranking system is bad.
3. If your ranking system uses a similar strictness of SBMM in social and ranked, your ranking system is bad.

Exactly. People don’t play in socials the way they play in ranked, so why base my ranked mmr off casual performance? At most, the social mmr should stop having any impact as soon as I finish my first placement. For all subsequent ranked games, you now have a mmr number for ranked that you can continue to refine as more data are gathered. Doing otherwise is just bad design.

Ditto with using hidden mmr to matchmake; in that case just make that number the visible rank number. And why even have socials if they are going to be matchmade exactly like ranked games? Pretty sure it defeats the point of even having a ranked mode if your “casual” mode is equally sweaty.

Really what they should do is overhaul the multiplayer menu so you can select social or ranked, then have the full set of playlists under each, with the difference being that socials don’t have ranks attached and the sbmm is looser. That way, people who want to sweat can go to ranked and socials can actually become casual. You could also have stuff like persistent lobbies and proxy chat in the social playlists, so that they are actually social.