Either rethink or completely revert the new ELO system

Open

artificialair67 messagesstarted Sep 22, 2026, 8:09 PM

other

artificialairOP2 days ago

I don't think I've seen a single person, top player or not, who thinks this system is an improvement in any way. People who were top 100 players before the reset are trapped in lower ranks and forced to effectively smurf for 20+ games to get even close to where players of their skill level are "intended to be." The general range of existing ELOs has also lowered drastically, putting people who used to be in middle to high Amethyst in a spot where they can now consistently queue players in the top 50.

I averaged around Amethyst 2 last season, and I am now on an 1/11 W/L because I physically cannot queue people in my skill level. Instead, because I started the season a little higher, I'm queueing people who were easily in Diamond / Guardian last season and am losing less than 10 ELO a game, so I can't even realistically rank down to escape this.

I don't know enough about the Glicko system to exactly identify the problem (especially because this specific implementation is entirely and intentionally undocumented) or suggest a specific change, but this needs to be seriously looked at. My shot-in-the-dark guess is that the "volatility" stat decreases at a much higher rate than it increases, if it's even able to increase at all.

211

artificialairOP2 days ago

ragesuggestions are the best kind of suggestions change my mind

Mercury | MudBrickWall2 days ago

well tbf all suggestions come from a place of rage

Mercury | MudBrickWall2 days ago

i think you need to be a very hateful person to play draftout at all

GracefulBeauty2 days ago(edited)

i think the system is just overtuned right now, it's way too sensitive to your first few games and then you get locked in place way too fast, so it takes a long time to correct a misplacement

GracefulBeauty2 days ago(edited)

your first few games are too volatile (because of queue rng, luck, and just how well you/your opponent happen to be playing in those games in particular), to have such a huge influence

GracefulBeauty2 days ago

the nice thing about the system is that once a player's rating settles, you're much less likely to eg queue diamond 3 snakezy and lose 30 elo to them like last season

GracefulBeauty2 days ago(edited)

because they wouldn't be diamond 3, just because they go on a losing streak doesn't mean they're suddenly not a warden-level player, and glicko reflects that better

GracefulBeauty2 days ago

as a concrete example though

GracefulBeauty2 days ago

@MattB8zer placed 1438 rating which is #53 right now after playing and winning only 3 games, against players who were 1122, 1288, and 1220 elo at the time

GracefulBeauty2 days ago(edited)

maybe they are a 1438-level player, but the info the system has so far absolutely doesn't justify such a high placement. they ended last era in amethyst 3

artificialairOP2 days ago

i agree with uh. all of that yeah

artificialairOP2 days ago

retuning falls under "rethink"
honestly the glicko system in general is probably a better fit we just need like. some room to move around man

sugaarush2 days ago

yea true

Vei2 days ago

yeah it’s sensitive af at first

Vei2 days ago

placed at 1250 and lost a game 12-13 to someone 20 elo lower and lost 90 elo or something stupid

Vei2 days ago

that was fun

sugaarush2 days ago

SAME LOL

sugaarush2 days ago

i think the part that needs to be fixed is that once ur elo stabilizes, there is literally no movement in ur elo relative to the elo tiers

Vei2 days ago

yeah -77 due to my trial chamber not having wax

it’s chill tho

Vei2 days ago

yeah I agree

sugaarush2 days ago

it doesnt make sense at all that u only gain 10 elo from a 30 min game (on average); when each tier is 300 elo

sugaarush2 days ago

even with a 75% win rate u need to win 60 games (30 hrs) just to advance one elo tier

sugaarush2 days ago

which is especially a problem when most ppl don't play competitive bc each game takes so long

MattB8zer2 days ago

although I do agree I think I am an outlier due to very low sample size. As stated just after, the next games will matter a lot more than a normal game and that is just similar to how chess does it, so I think its okay in most sense. I can see how it would be frustrating though.

Threadfinn2 days ago

There's a bunch of other people in the top 50 with a win rate of 3-0 or 4-0 against 1200-1350 players.

Threadfinn2 days ago

If you look at the win rate tab instead of the ranking tab on the website most of the top people are top 100 after just doing placements

Threadfinn2 days ago

Like here for example, yes they won against other very good players but it still seems like only 6 games is enough to just get lucky a couple times and almost are top 10 off that

Threadfinn2 days ago

Or these two farming pablomaker to get top 40

sugaarush2 days ago

also why did we remove the old elo system /genq

MattB8zer2 days ago

I think Im biased because you are just simply correct but Im still happy and dont want this taken away from me

pancelor2 days ago

huh, really? do you have some data or a simulation or something to back this up?

GracefulBeauty2 days ago

i imagine that's assuming you always play someone at your exact elo

GracefulBeauty2 days ago(edited)

so if it's +-12 per win/loss, you would gain on avg 6 elo per game at that winrate

GracefulBeauty2 days ago

so 60 games would be +360 elo

pancelor2 days ago(edited)

"if it's +-12 per win/loss" where is this coming from? surely glicko-2 is more complicated than that.
edit: I just read the example docs (https://www.glicko.net/glicko/glicko2.pdf) and 1. ok whoa yes its complicated, and 2. I don't understand why/how draftout updates your elo after each match. it doesn't seem like glicko-2 is meant to work on an individual-match level

Hursty452 days ago

I do think it was good at the start of season with the larger changes earlier but I do agree that like it should be more than +-12 on avg later on

GracefulBeautyyesterday

there's a minimum RD and i've definitely hit it personally

GracefulBeautyyesterday

it's about 10-12 vs same elo i think

GracefulBeautyyesterday

rating change depends on opponent's RD too though so it's complicated

panceloryesterday

ah, I see

panceloryesterday

In the old system, playing someone at your same rating would give the winner +30, right?

panceloryesterday

but in the new system, playing someone at your same rating doesn't give much evidence either way of who's better, so the elo gains are small

panceloryesterday

seems like the old system rewarded playing tons of games, and the new system is more concerned with finding accurate ratings. (does that sound accurate?)

panceloryesterday

maybe it'd be nice for the matchmaking system to not try so hard to get people at the same level as you. (or, that would be nice if you mainly care about your rating. the actual games are probably more enjoyable versus people at your same rating)

Douglasyesterday

I think an easy fix is to hide ratings for longer, and make people play 10+ placement games. Maybe they're still getting +70 and -80 behind the scenes, but we're not told what our rating is until it settles down.

Douglasyesterday

Once ratings settle as the majority of players have played 20+ games, the impact of 1 game goes down (compared to standard ELO system), and any significant movement up/down in rating is due to skill, not luck.

Douglasyesterday

It sucks to have an inflated rating due to a run of good luck, leading to one-sided (losing) matches, but that's a temporary issue, and is less likely to happen to a player getting their rating now vs. in the days right after the rating reset

artificialairOPyesterday

this isn't the only issue though
i should not be able to lose 10 games in a row and be in the same rank after the fact
if i lose 10 games in a row, the rank i'm in is above my skill level, i should not be consistently queueing the same people

artificialairOPyesterday

when it takes multiple weeks of playing for your elo to reflect your skill you end up with a large amount of people at any given time who are all over the place on the lb

Douglasyesterday

That's only happened to you in this case because you started playing right when the rank reset happened. Your rank was set early on when you had some "good" luck in playing opponents that had inflated ratings, whom you were able to beat. Eventually, everyone else's ranks started to settle, and you found yourself no longer surrounded by opponents with inflated ratings.

sugaarushyesterday

true, and vise versa

Douglasyesterday

Now that everyone's rankings are settling, someone new playing their placement games will have opponents whose skill better matches their rank, and won't have the same issue. This isn't a widespread issue (if it was, you'd still be surrounded by players of your own/lower skill that you could beat)

sugaarushyesterday(edited)

it works the other way around too tho; if i had unlucky games at the start, or say i learnt a bunch of tech and im better now, it feels like theres no movement in elo

sugaarushyesterday

cause each game is worth so little

sugaarushyesterday

to me thats the biggest issue

Douglasyesterday

This feels like a "grass is always greener" scenario. Under the old system, there were complaints that there was too much ELO change per game. Now there's complaints there's too little rating change per game.

Douglasyesterday

I like the new system because it means luck is less of a factor in the long run, and it means my games are consistently against players of my skill (instead of 2 lucky games leading to my next game being an automatic loss, and 2 unlucky games leading to an automatic win)

Threadfinnyesterday

I think the best, although maybe difficult to implement well, would be if the elo system stays the same but matchmaking gets adjusted based on recent win rate.

So if you're sitting at a 50% win rate alternating winning and losing, then your opponent will be about the same elo as you. But if you go on a win/loss streak then your queue range expands in that direction, and you can gain/lose more elo per game and are probably playing against opponents that better match your skill level.

Douglasyesterday

If the rating system is reliable, then matchmaking can continue to optimise for "2 players almost equal in ability, based on rating"

Threadfinnyesterday

The issue with that is that if someone improves suddenly, it will take a long time for their rating to catch up. All their games against equal rated players until that point will be unequal in skill.

Threadfinnyesterday

As graceful said, 60 games at 75% win rate would be 300-360 elo. Once you've gained that much your win rate will probably start dropping closer to 50, but it'll take 30+ hours of playing to be against equal skill opponents.

Douglas20 hours ago

Can you think of any, even hypothetical, example of "someone that improves suddenly"? The best I can think of is a new tech discovery, but that will improve the quality of your opponents as well.

An individual improving in skill takes time, which will naturally come with a gain in rating (either by playing multiple games, or by the in-built time-sensitive aspect of the Glicko system)

I haven’t lost faith in the rating system, but if you have, you should try to fix the rating system, and not push the problem onto the matchmaking system

Threadfinn19 hours ago

Personally my skill is highly dependent on whether I feel tired or not. I can usually predict whether I'll win or lose before I even queue. All I would have to do to get 200 elo better is not queue comp when I expect to lose.

marin6 hours ago

we are discussing the system and we will be making changes over the next few updates based on these suggestions

marin6 hours ago

personally, i agree that the rating change gets too low after ~20 games

marin6 hours ago(edited)

and i think it works well for getting players ranked initially (higher/lower than elo system), but with low rating changes after, it gets weird