Since I have started playing the game nearly four years ago the 1v1 ladder has used the same matchmaking setup. You can play people within 15% of your rating, both above and below.

What this does not mean: Buns has a 2292 rating so can play people above 1948.

What this means: There are 897 people on the ladder right now with ratings. 897*.15=134.55

Buns can play people up to "rank" 134. Rank in this context means the people with the 134 highest ratings, not the actual ranks you see on the ladder, ie. people not currently joined influence this, so do people with fewer than 20 games.

Buns can currently play ratings as low as 1841. The guy with 1841 can play anyone higher than 1700.

This was practical back when the ladder had few players on it. It isn't anymore.

What's far more practical is to drop that 15% to 5%, let Buns play the top 5% of the ladder, that's still 45 different people, specifically everyone with a rating over 1984.

The guy at 1984 can then play anyone over 1905, etc.

You would then start at the bottom of the ladder when starting, and climb, slower, as you wouldn't immediately play people with high ratings, you'd have to earn the right to play them by slowly increasing your rating, and thus the range of people you could play. It'd make it easier to push for truly high ratings on the top-end if you are skilled because you wouldn't be playing people 500 rating below yourself in lose-lose games (where winning the game hurts you and losing it completely wrecks you).

If 5% is considered too small, 10% would be an improvement as well. That would allow buns to play anyone over 1905, and someone 1905 to play anyone over 1791.

Right now some people already force the ladder to work this way by leaving when they can't play someone within the ratings range they wish to play. Making it work this way by default would level the playing field and act as a disincentive for such gaming.

Edited 8/6/2016 03:04:54

894 0.o

897, they can all be seen here: http://data.warlight.net/Data/BayeseloLog0.txt

And if people thnk this problem is hypothetical, here's a breakdown of the rating of Buns's recent opponents. His unexpired games break down to:

1800-1849: 3 people
1850-1899: 4 people
1900-1949: 7 people
1950-1999: 8 people
2000-2049: 2 people
2050-2099: 4 people
2100-2149: 2 people
2150+: 3 people

22 of his 33 active games are vs. people under 2000 rating. His average opponent has a rating of 1980.

There are 38 people with ratings over 2,000 right now. The matchmaking does not need to be this loose.

10% eliminates 7 of those games from existing, 5% eliminates 19 of them.

Edited 8/6/2016 02:52:52

+1

Fizzer requested I make a uservoice for the request when I mailed him with a link to this thread, the uservoice has been created and can be found here: https://warlight.uservoice.com/forums/77051-warlight-features/suggestions/15565602-reduce-the-matchmaking-range

similar request for RT ladder

All I see is:

"There are currently 383 players in the ladder."

The 897 are the number of players that have finished at least 1 game in the last 5 months, which is something completely different.

Additionally, part of your proof isn't correct. You do the breakdown of the rating of his opponents, but this is their current rating, not their rating when they were matched up with him. It is my experience (and note that I've been on the ladder for 5 years in a row now, always playing 5 games and mostly rated above 1800), that the difference when the games are made are mostly smaller than that and that differences over 200 elo points are rare.

Note that you are always matched up with the one closest to you in rating. So even if someone near the back of the range is available, you are still more likely to be matched up with the one closer to you. Since the 1v1 has a lot of players and a lot of games finishing at the time, it's rather rare to get someone close to the end of the rating. Surrendering at busy hours reduces that risk even more.

It is a problem though, but the 1v1 ladder really hardly suffers from this. Where it becomes really apparent is the 2v2 and 3v3 where there are much fewer teams and games. Ruil & Mohikanin really had this problem when they were first. Because of their large number of games, many opponents didn't fulfill the criterion anymore and they started playing 1600 teams while they were first. 3v3: a disaster with hardly any teams. We (me, me2 and Jackie) got a new team (no games finished) while ranked first.

What really needs a fix is the RT ladder as ps suggests. There is no matching algortithm there (based on rating) and the best player could potentially play the worst player there.

Note that TrueSkill would also solve this partially for 1v1, 2v2 and 3v3 as I've repeated again and again. A game against a low ranked team would always give you a mean boost and rarely gives you a small variance increase to offset the mean (rating = mean - 3*sqrt(variance)). BayesElo is a lot less straightforward in this. Winning against a lower ranked team could potentially drag you down. (TrueSkill is also easier to understand intuitively, BayesElo isn't that intuitively obvious.)
((Complete side-note to not derail the topic: in the RT ladder, TrueSkill currently doesn't work as well as it could because playing someone over 1000 points lower creates a neglible mean boost and a slightly larger variance boost. Which makes sense really if you use a sports analogy: do you think a first division team deserves a better rating for beating 6th division team? If anything, you know less about them, because "last week" they still beat another division 1 team, "this week" they only beat a bunch of amateurs.))

but the 1v1 ladder really hardly suffers from this

What do you mean MathWolf, the 1v1 ladder matchmaking is a complete disaster! One of the resones I left the ladder, was due to the knowlege that with my <2150 rating, I was still very, very far behind #1. Sine the wins I (potentially) would get was not helping at all. And odds of facing an above 2k feelt very small. (Keep in mind 2k, is still 150 points lower...)

That being said, I did play 4 of the then current top ten when I was myself among top 10. So it is not all "terrible". But I would like to see a few changes.Most importantly, only pair players once a day. I never understood the "need" to make the matches instantly. Paring once a day allowes players closer in rank to meet more often, and should reduce the extreme cases.

Also, @MathWolf, ranked players does indeed refer to players with a rating. Regardless of whether they have signed up for games or not. http://data.warlight.net/Data/BayeseloLog0.txthttp://data.warlight.net/Data/BayeseloLog0.txt is thus listing all ranked players. This is irrelevant to how many are currently participation. I did the same thing that Omniscient appears to have done, and ran the elo tool locally when I was intop 10. After playing around with it, it was very evident that playing an beating rank 134 is pointless. Top 10 needs to play more often. I mean one game between to players don't necessarily show a skill difference. But if #1 beats #10 7-1 that is a clear difference. (One of the cool things with Bayeselo is precisely that it cares more about beating the same opponent many times, while typical elo tools gives a flat + deepening on opponents elo only.

Rank Name                                     Elo    +    - games score oppo. draws
   1 Buns157                                 2294  167  131    31   84%  1981    0%
   2 ACLTears                                2272  212  164    16   81%  2007    0%
   3 Quicksand                               2260  161  127    38   87%  1873    0%
   4 alhazi                                  2231  127  106    48   83%  1903    0%
   5 Aldehyde                                2231  221  171    15   80%  1976    0%
   6 GiantFrog                               2210  260  192    10   80%  1981    0%
   7 HotBeachBum                             2190   60   55   199   83%  1866    0%
   8 TonySodano                              2174  196  165    18   78%  1893    0%
   9 TBest                                   2150  159  132    40   85%  1676    0%
  10 Roo                                     2131  490  274    12  100%  1421    0%

As you can see current (unoffical) top 10 has a very low avrg. opponent rating. (The +/- is how "confident" the system is in the rating of the player.

Edited 8/6/2016 11:37:16

Sounds more like a wish than a plea.

"There are currently 383 players in the ladder."

It may indeed use these people. Would possible explain why the wiki says matched within 30%, but I see no matches more than 15% or so different happening.

"Additionally, part of your proof isn't correct. You do the breakdown of the rating of his opponents, but this is their current rating, not their rating when they were matched up with him. It is my experience (and note that I've been on the ladder for 5 years in a row now, always playing 5 games and mostly rated above 1800), that the difference when the games are made are mostly smaller than that and that differences over 200 elo points are rare."

Technically right, it is current rating, but ratings don't fluctuate that much. He absolutely is NOT mostly facing people within 200 elo. I have not faced mostly people within 200 elo of myself. I respectfully don't consider your ladder experience super relevant. Sure, you're always on the ladder, but you finish around 100 games a year. You have finished 51 games this year. In a much smaller period of time than you have played 506 ladder games I have played 743 1v1 ladder games, and my experiences with the matchmaking were nothing like you describe.

Cherrypicking has been roughly as rampant as stalling. People have left and joined the ladder to try and force it to match them up against people closer to themselves in rating.

I agree with omniscient, the ladder matchmaking was created to try and help people get to their relevant elo quicker by having a wider range of players, but after you've gotten into your range, it's hard to move away from the average due to the wide range of choices.
and since the matchmaker is designed to try and get you to play everyone in your range once, then start over for the second game, etc *as applicable with games available* you will get alot of games that aren't really a test of your elo
the main problem I see with lowering the requirement is that it will make it harder for people to get out of the average range, and, people just entering the ladder will almost inevitably be stuck in the average even if they win every qualifying game.