I’m gonna write the quote again:
Stop strawmanning. I’m done responding if all youre gonna do is undermine any and all nuance for the sake of justifying your “take off the tinfoil hat” attitude.
Edit: Actually I’m done responding to you regardless of what you say. Its a ceaseless argument. Plus someone already said it earlier, but you are just baiting at this point lol
GDC. Josh Menke. TrueSkill2 paper. Microsoft.
It’s literally how TrueSkill2 is designed.
And Menke structured the divisions to match the standard deviations of the population curve. And that it’s structured for the 25:75 results.
But regardless. You can’t expect to rank up to Diamond just because you beat a team that had a Diamond player in it? Using that same logic you should rank down to Silver if the team you just lost to had a Silver player on their roster.
You can write it all you like. We are agreeing that the end point is the same.
I think it’s a natural evolution of an accurate ranking system. By ranking players accurately you naturally end up with predictable results.
You think it’s because 343 are lying and deliberately manipulating the result.
Yea the man inaccurately calls an exponential graph a linear one and then uses that incorrect linearity of kill count in the calculation that determines a players skill at the end of the game. Page 15. I might be misunderstanding it, but it seems like TrueSkill2 is a bad system for determining skill…
I’m confused. How is matchmaking supposed to place players against similarly able players on a small scale and place teams against similarly able teams on a larger scale if it lacks the faintest idea what to expect from these pairings?
I would think quantifying player skill, relying upon past data to form predictions, or expectations, and using this information to form matches would be the entire basis for skill based matchmaking. If you cannot predict what will happen before the match is played it would be a crapshoot.
Let me put it this way… I’d bet every single system ever created to match players based on skill has these type of expectations sitting somewhere in the wings, behind the scenes. Even if the information is not made public.
The game isn’t doing that, that’s why everyone hates it.
Yes, but most of those games dont have a monetized challenge system on a rotating schedule that is dependent upon RNG on top of that predictive ability.
Menke wasn’t an author on that paper.
And math’s isn’t my strong point (did get to Level 3 in Grade 12 and did some basic first year Uni maths - but dumbed down for the Medical Sciences).
But the graph of kill mean and variance look pretty linear. And they are described as scaling linearly.
The interesting part is the more exponential look to the death stats. That’s probably why they are have a reduced weighting compared to kills.
But the table on page 17 is pretty compelling.
I did misunderstand it then. Due mainly to my color blindness. I thought both the exponent parts were kill mean and variance and the linear ones were death mean and variance. Colors…
A statistical Number cant be correct. There are too many factors are not taken in account.
A few examples:
How did i kill a D4? Legit battle with same weapon or was it a lucky grenade or a power weapon?
Did i get the kill because i got the lucky headshot while my team mate got only an assist?
I was a plat 5 (few days agos, now plat 3, another story and evidence of this broken system). The next team i have to face D3, D5, Plat 1, Gold 5.
The system calculates that i should get 2 kill on D3, 1 kill on D5, 4 on Plat 1, 4 on Gold 5 = a total of 11 kills expected. What if 80% of the match im running into the D3 and D5? No chance to achieve 11 kills. What if 80% of the match im running into Plat 1 and the Gold 5. Probably i get more than the 11 kills.
Still, if im going positive, but 1 of my teammates, a gold 3 goes 3:14 and we lose, i get punished also.
So it isnt accurate at all.
This calculation sounds great on paper. It would work to some degree if all player in both teams would around +/- 100 CSR and the same ping. It would work if the personal performance in the match would decide how much CSR i win/lose. Instead its the team performance win/loss that matters more.
Well it could take in account my past matches in history like it does for slayer. Average of returned flags, assist on capturing the flag, capture / defend the base and so on. But we have the same problem as we have for the slayer statistics. The KDA of a match is not just Kills, Deaths, Assists.
Factors like how good / bad were my team mates, ping, desync, how do contribute a lot on the KDA outcome of a match. Next match - new game - new team mates - other opponents - another ping/latency setting - unpredictable outcome
Even if i play consistently (which is very difficult due to latency, desync, smurf accounts, and so on) it remains a statistical number of the past. It was build from matches with very different settings (ping, desync, rank of team mates and opponents, and so on)
And thats the point. It could work well with a big player base with a good spread of skill and csr. This would result in close matches. For me the average team csr shown at the end of the match is far less important than the reality. If the match is close, the matchmaking system did a good job. Both teams have similar skills. But if one team gets stomped than the matchmaing system failed.
I lost 70% of my matches in the last few days. Went down from Plat 5 to Plat 3. What does it means?
There are many factors that contribute to your rank. Mostly your raw skill of course. But yes, pings, equipment quality, time available, life stresses, alcohol habits, etc.
You can’t control all of them - but they all average out into your rank.
Your rank, K/D, KDA, average objective score are all retrospective scores that absorb all of those factors. And your KPM is the same.
It’s an average of what you’ve done before. Which is a pretty good estimate of what you are likely to do today. And Menke has commented how super accurate it can be - it pretty much matches to the number (after being rounded out to integers of course).
So if you can accept your K/D - you can accept your KPM.
And use it for what it is. A guide to where your performance is expected to sit. If you are falling behind then it’s an alert that you are off your game. If you are exceeding expectation then milk that run of form and try to get some wins and rank up.
Don’t get hung up on the exact number. If you are expected to get 10 kills and you only get 8 or 9 I doubt it’s going to destroy your rank. KPM is just a weighting after all.
Just keep an eye on the trends.
No ranking system is going to drill down in that detail. How would you even quantify the metrics. Or store the huge amounts of data?
It averages out. Lucky shots with unlucky misses. But with an increase in form the overall shots that hit go up.
So you dropped down bit over one division. A bit over 50 CSR points. That’s not evidence that the system is broken.
No. It’s just saying that in previous games against that quality of opponent you have scored around 11 kills.
It’s averaging out all the external factors.
The only thing it’s reading into your actual result is that you are doing better than before (12 or more kills) or not doing as well (10 or fewer kills).
Colour me surprised that a team game values team performance.
But this is why the system is so much faster to get an accurate ranking in FFA matches. Like it can nail you in 3 games from a 16 player FFA list. For a player out of a 4v4 list it stretches out to 46 games. And 91 for 8v8.
I haven’t seen a number put on current BTB at 12v12 - but it would have to be well over 100. Which probably explains the mixed experience people were having.
It all boils down to those things not increasing the accuracy of the system.
They are important for you rank - they contribute to the win.
But if you rank someone up on the basis of objective scores you just create a situation where they can’t compete, keep losing, and derank. And that would make people miserable.
Your KPM reflects your ability to win a 1v1 at that level. It can’t be manipulated (like KD or KDA). So the system can safely nudge your rank up knowing you can compete at the higher level.
Some stats, and I suspect it was assists mainly, were actually negative predictors. You should probably be ranked down if you are scoring well in this regard. And that hurts because I’ve always taken pride in chipping in with assists. But in some cases they reflect that you are struggling with winning 1v1 at this level.
They do.
But any ranking system is going to have to let that noise cancel itself out in the background. For most players it will be consistent.
Someone mentioned that they saw a post from 343 mentioning that ping was taken into account. If not for ranking then at least for matchmaking (eg. a weighting to support unfavourable ping). But I’ve never been able to put my hands on a reference.
Agree.
But the failure tends to be out of the match maker control. It can’t predict how a team gels. And unfortunately when population is stretched and teams are ranked wider - then the team dynamic can really break down.
It means, for whatever reason, you have dropped in form.
Plat 5 to 3 isn’t huge. A bit over 50 points. A drop of 1 and a bit ranks in Halo 2. Up to 2 ranks in Halo 3.
I did the same a few weeks back. Had a run of bad form (and poor pings). Went mid D2 down to mid P6. Took a couple of days off and then went to work at getting back up there. Pretty much took a day and half to get there.
You have control.
Did you stand back and look if you were doing anything wrong. Could you have arrested the slide? Were you getting frustrated and then making even more mistakes? (I know that I have to work at not pushing harder when I’m going negative in a game - because that invariably just leads to more deaths).
Did you need to just step away a couple of hours. Even professional sports people have bad days.
I feel SBMM doesn’t stand a chance not due to low population but to a varied population.
Ranked has four game modes; slayer, oddball, CTF and KotH. Players are not going to like all the game modes and will have different variants of skill for each one.
Overall this is a very niche playlist and has been in all previous Halo. Players tend to pick a game mode they prefer but Infinite they can’t do this. They have to struggle or excel through all of them.
Jack of all trades master of none!
Secondly progression is key to human satisfaction and the progression ticker in this ranked playlist doesn’t progress.
I believe the CSR should be decoupled from the division and you should in some way have two progression tickers. The first being the current CSR but just in a number form … the current system is fine 1220, 990, 1400 or what ever resemble your current personal skill. But each ranked playlist should also have a division assigned to you purely based on wins/losses with respect to ever other player - top 10% in Onyx etc.
I’m sure this way will have its draw back too but at least players will still have a progression that goes up and down more freely than what we have now.
I think the SBMM does ok. Most of the games are still close. Or at least could go either way.
The blowouts are frustrating though.
Your point about game modes is pertinent though. Maybe they should only assign objective games if the rank spread is narrow.
I just feel that Slayer is less likely to blow out if your team doesn’t gel.
1000%
I still can’t believe they didn’t put an XP progression system in.
It’s mind boggling.
I’m just not sure skill rank is the way to do it.
Tending to agree.
I would take the CSR number away entirely. Just have divisions and tiers.
Then everytime you win a game vs a closely matched team (and good ping) you get a championship point for the win. Announce at the beginning of the match that a point is up for grabs.
So during the season you collect points for your division. Have leaderboards for each division. And then across regions, countries, age groups, game modes, and international.
But we are talking about 70% matches lost. The team that was thrown together by the match making system lost this 70% of the matches. Most of the matches i went positive, but lot of team mates went negative. In some matches we were stomped. 0 KOTH captures, the other teams had either better players or were a stacked teams.
My form didnt drop. I cant perform how its predicted, when the whole other team has better player. Not in rank or csr, just better skilled. When the whole other team got positive k/d, but our team just struggles to get even k/d, my form is not responsible for win/loss. Sure im contributing to W/L but in the end its the team. The player were picked by the match making system.
It doesnt help, that we know that because of low plaver base, the skill of the teams can differ a lot.
The match making /SBMM cant work as intended when there are not enough player to build even skilled teams.
Like i said: i dont take a look at the player ranks or the average team csr. For me there is only one counter: was the match close or not. If it was close, i dont mind to lose. The other team got a better hand. The match making system worked as intended.
If 1 team gets stomped, dont matter if we or the other team, the match making system failed. Its that simple.
Which measures could be applied in such a situation “match making /SBMM wont work as intended due to low player base” is a whole other discussion.
That’s easy to answer
Its because the sbmm in casual is too strict in social combine that eith the other issues that infinite has with the big ones being dysync and the lack of content
You get placed into unfair matches filled with lag and dysync 85% of the time on off hours and 50% on prime time hours
And thats not to mention having to sweat for those matches
Its no wonder why the playerbase is falling off at a rapid pace
Like Darwi doesn’t want to admit it but the strict sbmm in casual is one of the causes of people leaving like everyone in my friend group and family who prefers to play infinite social mode over ranked lists
These three reasons
*dysync
*lack of content
*the sbmm being too strict in casual
And when i was testing out a few modes with a few mcc forgers we all were talking about how its nice that we can goof off in mcc matchmaking if we wanted to but when infinite was brought up which a majority of us said it’s comparable to being kicked in the balls due to how much you have to sweat in the casual modes on a game filled with dysync
A quick glance on Halo Tracker and for your last 25 games you are going 11W and 14L
That’s closer to 45%
So if you were talking about an early run - you’ve at least recovered.
Your last 35 matches include scores of -5, -3, 0, -2, -8, -2, -2, -8, -9, -1, -1, -4, -4, 0, -5, -6, -3, -7, -1, -4, -4, -1, -6, 0, -9, and -8.
You’ve only been positive in 9 out of your last 35 games?
They were. And you only have to base your performance off the expected.
If you play your part in the objective and meed your expectations you have probably done your bit.
The match maker can only match the teams on the basis of the individual’s skill combined.
How they gel as a team is in the lap of the Gods.
Unfortunately the bigger the spread of ranks the bigger this affects the result. ie. the occasional stomping.
And I agree. That’s not pleasant for anyone.
Not entirely. The ranking changes are based on the result and the relative ranks of the players/teams.
The stomping is irrelevant. To the system it is simply a win or a loss.
It only truly fails as a ranking venture if the underdog is thrown an easy win (eg. if someone quits).
And I get that is a harsh reality for someone desperately trying to rank up. It can feel like a waste of time. But unfortunately the low population is exposing this aspect of the system. And remember, it’s not just this system - any match maker system is reliant on a healthy population to choose from.
Again the key is how the system is performing overall.
It is still self checking itself. If similarly ranked players are performing similarly (50:50) and players apart are performing proportionally (eg. 75:25 at the division level) then it can be confident that the ranking is working.
Unlikely. Are more people playing custom game browser or matchmaking in MCC? Matchmaking, by a long shot.
Frustrations surrounding SBMM are valid, but I wouldn’t consider its mere existence “egregious” as posited by the OP. Every single multiplayer game has had this to one degree or another. Halo has had hidden systems since day one of Xbox LIVE.
The main problem is that its too strict in social like everyone that i talk to regarding the game lists it as one of the three main reasons why they don’t like playing infinite
Legit was playing with a few people on mcc and we had a kid join us to test out some modes and maps on mcc and we asked why he prefers playing mcc over something like fortnite and infinite
His answers were that he doesn’t like fortnite or battle royales and that infinite was too hard for him and that he didn’t like having to “try hard” every match every game of tdm
I agree; I think the parameters are too strict in Infinite and MCC. But I also do not work on these systems and nor do I have to worry about creating a matchmaking system that caters to millions of people, so I don’t want to pretend I have the answers or anything.