Open Pairing Options

Just curious to get some extra thoughts (while I'm deep in these weeds). 

Let's say we have 4 players who get matched for a game. 

Adam is rated 1200.
Becky is rated 1300.
Charlie is rated 1500. 
Daniel is rated 1600. 

How should they get paired? I'll show the options and you can vote. In each case I'll format like this: 

Team A(1) vs Team B(1)
Team A(2) vs Team B(2)

So one team is on the left, one is on the right, and you can see pairings. 

Option #1: Attempted Balanced Average 

Daniel(1600) vs Charlie(1500)
Adam(1200) vs Becky(1300)

Option #2: Closest Ratings Match

Daniel(1600) vs Becky(1300)
Charlie(1500) vs Adam(1200)

Option #3: Staggered

Daniel(1600) vs Charlie(1500) 
Becky(1300) vs Adam(1200)

Option #4: Random

? vs ?
? vs ?

Currently we do #1

I prefer #1.  It gives the best chance of a balanced game.  #2 and #3 happen all the time when two people specifically pair up, and it usually means a couple 1400 guys get massacred by a couple 2000s.

There is the downside that a 2100 guy can get saddled with a 900 against a pair of 1500s.  Since the average rating is 1500, the rating system will consider it an equal game, but the 900 is almost sure to commit suicide before the 2100 guy can take down his opponent.  But the only solution to that is to prevent that pairing in the first place with ratings-filtered seeks.

#1 and #3 are nice but #1 seems more equilibrated, considering the average strenght and not an individual radius like #3.

#2 is weird. Seems like the second team is in brutal disadvantage.

#4 could be an alternative option I think, for adventurers.

If it was up to me I'd keep #1 as standard and make #4 available just like live games, with an open rating range so it wouldn't exceed the maximum range for more/less.

Anyway, it's a bit confusing, since you must also prioritize getting people paired ASAP, so, maybe stick with #1 and follow up on improving it.

It could be an intercalation on #3 also, varying from 2 higher rated opponents / 2 lower rated opponents, and so on, eternally and forever, so no one gets impaired.

This is the standard for MOBA games, you get paired with weaker opponents, if you win, you're paired with better opponents, keep winning, better opponents, lose, lower rated opponents and so on.

It's like besides from your direct rating, by playing teamed up with another person it is justified to have an indirect rating, based on results (indirect variable) and not just the (direct) rating itself.

I don't think having an option between #1 and #4 makes sense.  Making it an option means you have to pair people who chose the same option, so you're splitting the pool of players into two groups, so you have to wait longer for games.  I think sticking with one option is the best way to go, and #1 looks best to me.

Perhaps the best way to improve pairings is to have rating-restricted seeks.  Presumably with a broader range than for normal two-player games, though, as while you don't generally want to play someone 300 points above or below your level in normal chess, a 1500 and a 2100 could be a roughly even match for a couple 1800s, and that's a 600 point gap from the 1500 to the 2100.

Rating-restricted seeks will make it take longer to get a game, but as the interface, etc., continues to improve, maybe some people will be tempted away from other sites.

I understand.

I agree.

#1 is good but is also #3, considering that higher/lower pairing I mentioned.

I know it seems a bit confusing but it's rather simple.

Nevermind.

#1 is the best by far.

#1 makes the most sense in general (when you have 4 players looking for random partners).

However, I think it would be a good idea to switch it up.  Rather than deciding which of the options to do 100% of the time, I suggest all of them be options with lopsided probabilities.

Additionally, high vs. low can be interesting sometimes.  By that I mean you pair the high guy with the low guy on a team and they play the two medium players, but the high guy on one team plays the low guy on the other team.

Option #4: Balanced Teams, High vs. Low Games

Daniel(1600) vs Becky(1300)
Adam(1200) vs Charlie(1500)

For example:
#1: 85% of the time

#2: 5% of the time

#3: 5% of the time

#4: 5% of the time

Or some other probabilities lopsided to the fairest matchup.

Another possibility:

#1: 70% of the time

#2: 10% of the time

#3: 10% of the time

#4: 10% of the time

#4 makes sense if, when, you randomly select 4 people from a queue, and they're within +/- (variant) RD of each other.

You can't disregard truly random.

But when ratings are skewed across a random selection of players in queue (or, those waiting the longest) ... it makes sense to implement #1, #2, or #3.

I've requested additional data points to try and weigh-in with what would be best for everyone. (Outside of allowing people to set parameters in settings at the expense of a longer wait time between games for this to be aggregated and computed.)

#1

On the subject of high vs. low, see the following conversation from the JannLeeCrazyhouse match.

FM chickencrossroad: switch
FM chickencrossroad: better for lag
eekarf: how switch
caspiwins: how can we switch?
eekarf: its high vs high
eekarf: automatically i think
playe4plz: lol what its always high vs high

As mentioned in another thread, I cannot see what JannLeeCrazyhouse himself said because I am following him -- as non-intuitive as that is.

This is an example where all of the players involved in the match want to play high vs. low, but there is no way to do it.  Their point is such a matchup can better balance the lag of the players with certain configurations, such as high vs. low in this case.  Also the playing styles can be such that the players will prefer high vs. low -- for example, some people trade more than others, and players may prefer a high trade board to a low trade board.

Generally with high vs. low with large disparities in ratings you get a situation where the high player on both boards can beat up on the low player, so it becomes a race to see who will mate first.  That is sometimes more interesting for attackers than 2 balanced games.

Option 1 is best.

I would prefer to wait slightly longer to have 4 players with more similar ratings.

Yes, option one divides the teams up evenly, but perhaps it should randomly switch between high vs high and high vs low.  Using the original example, that would mean half the time it would do this:

Daniel(1600) vs Charlie(1500)
Adam(1200) vs Becky(1300)

and the other half it would do this:

Daniel(1600) vs Becky(1300)
Adam(1200) vs Charlie(1500)

Both ways the teams are roughly even, but sometimes you're roughly balanced on both boards, and sometimes you have a race to beat the minnow.

I think that the selection criteria among several options should be weighed and balanced by overall queue activity ... in whatever order a queue is managed. (Give me data, or give me death.)

In the random pool I think high vs. low should be thrown in the mix, but half/half (50/50) is a bit much.  High vs. high games are generally better, with some exceptions as noted earlier.  I think a better ratio for hvh/hvl would be 80/20 or 75/25.

However, I think it's important to let people manually match however they want.  If I am the high player and I manually match the low player on the other team, then it ought to play hvl every rematch.  If I am the high player and I match the high player on the other team, then it ought to play hvh every rematch.

I wouldn't write those conditions into software. You know, maybe. But only if i had the time and it was a side project after all other considerations are handled.

In my (OTB) experience ... people at least swap positions. Sometimes both sides (simple end-game piece swap), or one pair stands-up and switches places.

HvL or HvH every rematch (assuming same teams) gets old and boring and not that fun pretty quickly. (Why bother to have teams, just go play ZH if this is the way you want to go, or simul yourself, etc.)

Accommodating every extraneous or specialty condition, whims, wants or desires outside of the norm or some standardized system is the fastest way to make a mess of software and make it more difficult to handle when someone else comes up with the next great idea and requires it be integrated.

In my (decades long) OTB and online experience, people generally play a long series of high vs. high, then at some point say "switch" and play another long series of high vs. low.  You can see an example of that just hours ago in the JannLeeCrazyhouse matchup.

The disadvantage of switching every game is it's harder to get a feel for the opening.  People generally feel their opponents out in the opening by playing the same person (often the same color OTB, or switching regularly online) over and over to figure out holes in the opponent's opening.  Constantly playing a different opponent makes that impossible, which is very annoying.

Playing OTB with my family, we usually do HvH, infrequently HvL.  We switch colours, but not usually positions.

But we often have 5-6 people playing, so we tend to rotate out after 2-3 games anyway.

Color illustrates another problem with a rigorous 50/50 hvh/hvl switching.  If you switch color every game and switch opponent every game and play the same team in a series, then you would in theory always play white against one opponent and always play black against the other opponent.

In chess.com's case that probably wouldn't happen because there seems to be no rhyme or reason to color.  I constantly hear complaints from my partner of "why have I been white 5 games in a row?!" or "why have I been black 5 games in a row?"  (I mention my partner rather than myself because I don't really pay much attention to the color... but I do remember my partner's chat.)