I was told its mainly based off of KD. I have lost many games and not dropped once. I play with friends who happen to rank much lower than me (not trying to get easy lobbies just playing with friends and it puts us in my lobbies anyways as they usually can’t compete that well). When we win i go up in huge chunks and every time my KD has been very high and my overall score is usually the highest in the lobby for both teams. So i would say if you know you are losing the game at least go for a positive KD. That is until the update and who knows what will happen then.
Obviously you just can’t trust anyone
I think he puts more weighting on K:D than I do. But that just reflects that we don’t know all the metrics used or with what weights they are applied. But also keep in mind that he has done more deep dives into player game histories than I have.
My feel is that kills/min is far more important, and that it’s effect is around placement (when your curve is wide) and that it ranks you faster not higher. Particularly as it gets harder to have really high kill rates vs good opposition.
And the paper also mentioned that weightings around kills or deaths were largely reflected in the result itself, so not all that useful.
We are also hoping that any such K/D weightings are softened for objective games. That’s certainly implied in the MS research paper - but it’s not clear what’s happening in the wild.
I would love to have access to the stats API. We could easily set up a page that tracks your last ‘x’ number of games and looks for trends in K:D around changes in CSR.
The player experience is paramount.
Which is why they need an XP rank as soon as possible.
But rank doesn’t (and shouldn’t) work that way. You rank up by getting demonstrably better at the game. Not just by winning.
https://www.xbox.com/play/media/EGF6F79H
This is the player experience in ranked
-
The K/D thing was just an example to explain my point, the differences will be subtle expectations around the entire performance. We aren’t fully aware how MMR is calculated so in terms of your performance, the same statistics will hold different value with certain teammates and Vs certain enemies. Either way let’s say it takes everything into equation like score, wins, k/d, and gives you determined value X + 2. Your teammates value is X + 1 which may be only slightly lower than you. As a result of your lower teammate you are facing opponents rated X + 1.5. So if you are matching performance against X + 1.5 players the system will think you should be closer to 1.5 than to 2. Your MMR will actively fall unless your performance is higher, as more is expected in this scenario from you for being the highest ranked player. Your friends are close on rank so expectations will be minor, but they still exist and factor into how your MMR is calculated.
-
Winning is important, but we don’t know how important and we also don’t know how much it offsets your MMR. Extreme scenario, but imagine a player in Onyx who suddenly starts going 1-15 every game but still winning due to luck. Their MMR should be dropping rapidly to get them in more appropriate games. Expected win rate surely plays a part in it too, I can’t remember exact numbers but I know teamed players are expected to win 52% of all games and solo players 47% so as a team you are already expected to win more games. Losses probably hit harder due to that fact also. But yeah you may not be losing MMR on losses, but in your case we know for certainty that your MMR value is lower than your CSR based on how your rank is moving. This could be because on the wins, you are already playing with lower MMR rated players, so at this point your MMR could go up, but not enough to cover the current gap in your CSR. It could in extreme cases be going down slightly too. So even with wins you are inching further away from your current rank. I can’t say if you are or are not gaining MMR on a game, we just don’t know. Ultimately there will come a point where your friends and you will lose similar CSR as you gain as your MMR levels out with your CSR. If CSR falls below MMR, you will start ranking up quicker than you fall. However, that’s less comforting when you’ve reached a peak as you just want to be back at that peak again. Although for most players that is unrealistic and they will hang around that middle ground which is closer to their actual skill level.
-
Your MMR may be lower before you even start playing ranked. I have no evidence it’s in Infinite but rank decay exists where as you aren’t playing your MMR lowers over time. Your MMR can be affected by social too. Lots of unknowns I know. Darwi and I both have beliefs as to what is the most important thing that affects your MMR but neither of us know for sure and I’m glad we don’t. It is likely that with the rank reset the performance and MMR calculations are modified. They have lots of data now to use to see how different stats affect win rate.
-
The weaker opponents would be in terms of MMR, that is something I can say with certainty based on how your CSR is behaving. It’s hard to take an enemies CSR at face value as you could be playing other players who are freefalling. Sometimes after the game you see a player this is happening to with a giant red bar even with a good performance in that individual game. Rough times. If you and your squad felt that the games weren’t getting easier then that would suggest your MMR is still probably evening out. If the games do get easy you’ve fallen too far and then you start bashing skulls and going back up. You won’t freefall forever, but unfortunately you may also not return to the same peak you once knew. That’s a very tilting aspect. I was 1800 once, why can’t I be it again?
-
I get that and I was obsessed with my rank in Halo 3. I wanted to be a 50 in everything and then not play as all I cared about was the status, not the games. However, now I just want good games. I don’t care if I’m Gold 2 if it means I can get a lot of good games that finish 50-49. It’s nice to see the rank go up though. I’m starting to hit my plateau now too at the 1650 mark. Soon I’ll join you in the freefall. That part sucks but I like getting good games more than the ranking aspect. It’s not for everyone but my overall goal will still remain. I do want to rank up, but I’m in no rush to do so. When I develop and can handle the players at my current MMR more easily I’ll start to consistently rise again until I hit a new barrier. But the whole way through I’ll enjoy fighting tooth and nail to swing as many of those 50/50 games in my favour and steal CSR from the enemy team.
If Halotracker doesn’t show a % those are 50/50 games based on all players current MMR.
2 Likes
They need to have a breakdown of CSR after each match.
They also need to strive to make matches more even so less instances of one team “should win” (at least by large margin). This is the main reason for swings in CSR. If you’re supposed to win and win, you get little CSR. If you’re supposed to win and lose, you lose a lot. My issue is, I don’t trust the system predicting the “supposed to” part. And even if it was accurate, who wants to take that bet?
Imagine going into a casino and placing a 100 bet. If you win, you make 10 but if you lose, you lose the 100. That casino would be out of business because no one would want to play.
Suggestion:
Winning:
Make a standard amount gain per win (let’s say 3-5 CSR) depending on who’s “supposed to win”
Add win streak bonus. (2-10 CSR)
Add personal performance bonus. (1-3 CSR)
Losing
Lose standard amount per game (3-5 CSR) depending on who’s “supposed to lose”
Add personal performance reduction so the bottom performer can lose 1-2 CSR extra
The key here is going on win streaks to gain rank. Winning needs to be valued above all else, especially personal K/D as that is the point of a arena shooter - to win.
*The amount of CSR was for reference only, and will have to vary depending on where you are in the ranking system.
All fair enough points, and I think my point was just about margins–at the end of the day, I really don’t think I played particularly poorly based on individual performance, wins, and opposing team strength from visible CSR. The fact that it’s taken us thousands of words of text to establish some goldilocks-esque theoretical reason why a 1.0 KD performance where we won all the games we were supposed to win and then some resulted in a rank free fall is just not a good thing.
And that doesn’t even get to why the system felt necessary to blast me up to 1582 anyway. That wasn’t a slow grind, I was consistently in 1530-1560 territory for weeks and then over a session or two it blasted me straight up.
The “CSR is different than MMR” thing feels like bad science in that it’s a tourniquet we can apply to our understanding whenever convenient. Game doesn’t make sense? Well, we can just say MMR was way off everyone’s CSR so that’s why–unprovable, unknowable, and unsatisfying to the player.
Anyway, we’ve talked this to death. The issue is, as pops up in this thread and all the others that are going to keep coming up every day–it’s not fun for many people not to be able to connect their games to their rank. I’m glad you seem unbothered by free falls and pitches and changes in CSR you don’t understand and can’t really pin down to anything. But I’m bothered enough to stop playing the game, and so are most in my circle, and so are a lot of people in this forum and on Reddit. We’ll see if 343 does anything about it. If not, there are lots of fun games out there, and I’ve probably played The Big Three (Streets/LiveFire/Recharge) about enough for a lifetime anyway.
1 Like
Halo 5 let you search for closer skill games, but the main issue was it took ages to find games most the time as you want a tight skill and connection match. Good for some but I’d rather role the dice and just get a game going as soon as possible.
The supposed to win team is something they calculate over millions of games. What team was expected to win? Did they win? They can now see all of that info of how many predictions were correct. Trueskill2 can predict the correct winner in 68% of games, which considering how it strives to get games as close to 50/50 as possible at any given moment, that’s a very high number which it can accurately predict the outcome of a match solely based on the players assigned skill level.
Makes me think MMR is highly accurate otherwise predictions would be far less reliable.
If players got CSR just for winning it would mean nothing if the teams were predetermined to win. A user here was stacking a full team of 16-1800+ Onyx players and stomping everyone. They barely gained any CSR and won 48/50 games or something mad. The two losses hit extremely hard because they had not increased their MMR at all. You are rolling a dice when you search, but I suppose you can only match who is online around your connection and skill level at the time. Won’t be perfect 24 hours of the day at every skill level and every region.
There is potential for a very aggressive system if this was the case. I wonder how people would feel about losing 30 CSR after going -8 in one game against much higher rated opponents, if it meant gaining 30 CSR after going +8 against much lower skilled opponents.
I like the system, but I see a lot of complaints. I’ve also seen complaints about every ranking system for every game I’ve ever played. So on one hand somr people will never be happy with how it works, just as long as too many players aren’t being deterred from playing.
Realistically it would help if they offered an incentive to play games aside from CSR gains. In Overwatch you get credits per win to spend on golden guns. At the end of the season you get bonus credits depending on your highest rank tier. It’s a good way to get people to play ranked, enjoy their wins more and I guess not sweat the loss in rank as much.
1 Like
They do not have millions of games played on individual players. Just aggregate. From personal experience, I can say the system isn’t very accurate. They will place two low diamonds with me and a friend in Onyx and “expect” us to beat a mix of high diamonds and low Onyx. This is a crap shoot at best. We are forced to play our buns off, carrying these lower ranked players just to be rewarded with 1 single CSR. But if we lose, minus 10. You can use whatever explanation you want, that simply isn’t an enjoyable experience.
Furthermore, If you read my post thoroughly, you will see that I do think a CSR range should be applied to those who win or lose games they are “supposed to win” but that range needs to shrink to reduce the swings that are clearly -Yoink!- off a lot of people.
More than anything, they need to implement a CSR breakdown screen so you can at least see why you gained or lost what you did. That would go a long way to reduce confusion/frustration.
Here’s a scenario for you to think about, as always numbers made up.
Team 1 - 2,800 MMR
Player A - 1,400 MMR (CSR 1300)
Player B - 1,400 MMR (CSR 1500)
Team 2 - 2,800 MMR
Player C - 1,400 MMR (CSR 1400)
Player D - 1,400 MMR (CSR 1400)
If team 1 loses, player A will lose minimal CSR as they are already 100 away from their MMR. Player B will lose a lot of CSR to bring them closer to their MMR.
If team 1 wins, player B will gain minimal CSR as their MMR is still far below their CSR and they are only beating players at this lower MMR. If they played players at 1500 MMR at this point, they would likely lose a lot/do poorly. Player A will gain a lot of CSR because their MMR is above their CSR.
If team 2 wins they gain some CSR, if they lose they lose some CSR. A fair amount.
How did Player B get to 1500 CSR in the first place? Maybe a bit of luck. Maybe their skill hasn’t grown in tandem with the rest of the community that has developed over the days/weeks. Maybe they had a good run of gametypes and maps they are good at but as these averaged out over time so did their skill. Bunch of reasons.
But that’s how it works, so you don’t really need to know an exact breakdown as you can infer your MMR by how your CSR behaves.
Again, I appreciate the detailed theory but at the end of the day, it simply isn’t fun being forced to carry players through very sweaty, difficult games with very little to gain from it and a lot to lose.
To speak to the accuracy, the reason they are doing a mid season reset tomorrow is precisely because the system isn’t accurate or behaving as intended. I’m sure you saw the Reddit thread detailing how you can manipulate the CSR you gain by throwing social matches and effectively lowering your MMR.
Add cheating, disconnects and the difference between a player skill when being carried vs playing solo and I think the system is probably hovering around a 50/50 at predicting the actual winner. Which is to say, it’s not doing much.
At the end of the day, the current system isn’t fun as reported by countless players. It isn’t accurate as made evident by the mid season reset and it clearly needs a rework.
1 Like
That’s not true, the Reddit poster either misunderstood or intentionally misled people. That manipulation would not work in any capacity. You don’t want to lower your MMR for any reason other than smurfing. That strategy cannot be used by someone at their peak CSR to gain more CSR. It will result in them gaining minimal CSR for wins and losing massive amounts of CSR for any losses when they do lose. This is because CSR wants to be equal with MMR.
They are changing the system because it is “too generous”. Most players will find themselves at a lower rank after the reset. They may change some of the metrics to measure performance, but this will not change the relationship between MMR and CSR. If your matches are sweaty, and you lose then lose lots of CSR, this means you’re losing to teams rated worse than your visible rank. If you’re finding it hard then naturally your visible rank is too high for you at this point in time as otherwise you’d easily stomp these lower rated teams.
The relationship between MMR and CSR is a pendulum. It wont always swing in your favour but if you stray from the centre it will guide you back towards the middle whether you are low or high. Once your skill improves, so will the point your rank marks as the middle point. You don’t consider the lowest point of the pendulum to be your true rank, so why should the highest point?
1 Like
You may be correct about the MMR vs CSR. I’m just going of what I have read.
In any case, if your number of 68% accuracy for the TrueSkill2 system is correct, we should only see swings in the 32% range.
For example, using 100 CSR as the base:
If you are supposed to win with a 68% probability and do, you should expect to be rewarded with 68 CSR vs 100 CSR if it was 50/50.
Likewise, if you lose, you should only lose 132 CSR vs 100 if it was a 50/50.
From personal experience, I am routinely gaining 1 CSR from a win and losing 10 for a loss. Using the arbitrary scale I provided, that would mean gaining 100 CSR for a win you were supposed to win and losing 1000 CSR for a loss you were supposed to win.
The difference between the accuracy in which the system can predict outcomes and the variance in the reward system is too great. That is where the grievances are coming from
The 68% is not the probability of you winning, it is the probability of the system being able to predict the outcome of the match based on the players involved in the match/ Out of every 100 games, the system accurately predicts the winning team in 68 of them. Trueskill1 could only predict with 49% I think it was. So, this means the model is more advanced and is more accurate at determining MMR. If the MMR of players was not highly accurate in most cases the system wouldn’t be able to properly use data to determine the winner so frequently. As games are matched to be as close to 50/50 as possible, this makes the system predictions even more impressive.
You are losing 10 for a loss and gaining 1 for a win right now. But that won’t be the case when your CSR and MMR are close together. Then you are more likely to see 5-10 for a win and 5-10 for a loss. If your CSR drops too much you will see -1 CSR for a loss and +10 CSR for a win. Unfortunately for you, and other players who are frustrated by this, your CSR is too high for your MMR right now. There is nothing any system can do to fix this as your MMR is reflective of your true rank, if you were matched on visible rank you’d have an awful time once your CSR got too high for your ability.
When I was in the 1500s for Onyx I lost 12 games, won 8 and ended up almost 50 CSR higher than when I started. Now I in the mid 1650+ range my wins and losses are lower than before. Based on my gains and losses I’d predict my MMR is still a little higher in the later 1600-early 1700 range. Once I hit that it will be incredibly hard for me to rank up and much easier for me to get back to the rank I am supposed to be. Let’s say it’s 1675. If I fall below, it will always be easier to get back to it, if I go above, it will always be harder to maintain it and I’ll fall quicker. Once I get better I’ll be able to get higher and stay there consistently, but I could also get worse and end up stuck in the low 1600s for a while.
1 Like
I disagree with the idea of MMR I certainly don’t think it represents your “True” rank.
MMR is a compilation of all your game data from all game modes, including social and BTB.
I use a lot of those modes to mess around with friends (sometimes have a few drinks too lol).
Why should that, unranked, game data have any bearing on my ranked performance which myself (and nearly every other person on earth) treat very differently from non ranked game modes?
The difference in your ranked CSR vs the opponent ranked CSR should be the only thing that matters. Otherwise I’m stressing about how my MMR is impacting my ranked performance. As you said earlier, doing poorly in unranked will negatively impact the amount of CSR you will gain in ranked. To me, that completely defeats the purpose of fun, mess-around unranked playlists. It makes the whole game a sweat.
But if your MMR is inaccurate your ranked games are much easier, all games will be if you deviate from trying to not trying. Much easier, you stomp, MMR back up, CSR up.
MMR is your truest rank. Visible rank matchmaking promotes smurfing, deranking and would lead to poorly balanced games even more frequently due to luck. We basically had that with Halo 3. The ranked matchmaking gets much closer and better games consistently now even if it doesn’t feel as good as a rank.
Yes you stomp, until you don’t because, like you said, they can only predict to a 68% accuracy. So when you do lose, you lose 10x And when you win, you get next to nothing. Not a fun game loop.
There’s no reasonable justification why unranked play SHOULD impact your ranked play, other than smurfing. There are way better ways to deal with smurfing that don’t ruin the relaxing separation of ranked and unranked. Like when they perform really well in a low rank, they should gain a lot of CSR. But you don’t need an MMR to do that.
That decision simply makes the game less fun.
Personally frustrated with the system myself. Yesterday after regaining my Diamond 5 status I had 1403 csr. I ended up 1401 csr after 10 wins and 8 losses. I lose like 12 points for each loss while only gaining 1 to 6 points for wins. Also note I lost 12 points in two matches despite them being 3vs4. I have to win 4 games to make up for that. That isn’t fun at all. Doesn’t matter what rank I’m at. I could be Bronze 6 trying to get to Silver 1. Having to have a 67% win record to rank up isn’t fun. I suppose I just need to have amazing performances despite quitters or the severely imbalance teams. Why are Platinum players playing way above their skill level anyway? They play better than Diamond 5 players.
I won 12/15 games the other night and end up down slightly (Onyx 1600). Lol how in the world does that encourage engagement? I needed to take a few night off after that.