Guild Wars 2

Was debating making this topic since everyone and their cousin has a solution to the problem of matchmaking and there is no solution that everyone will be satisfied with.

There’s a lot of controversy over the use of glicko since it was never designed for how it is being used, but I don’t think the issue is with the glicko itself but the matchmaking

Right now glicko causes a couple things:

• Glicko walls/hell where servers are locked in certain tiers
• Stale matchups where the same 3 servers are battling it out every single week

This is because the matchups are based on glicko, with the reasonable (but wrong) idea that servers of similar glicko scores will be similarly matched. But, because of score differences between divisions, a server that is underperforming or overperforming may have no chance to move down/up to get a better matchup.

So the actual idea is we have 4 tiers. Keep glicko as it is, but at the end of every match, look at the glicko scores within each tier. If a server has the most glicko points at the end of the match (regardless of whether or not they won), move them up a tier. If they have the lowest glicko, move them down.

Each new match would be the middle-glicko from the last matchup, the lowest glicko from the tier above it, and the highest from the tier below it. This wouldn’t really help T1 and T4 except to give them different matchups, but they’re not getting anything better now.

This eliminates both the stale matchup and glicko wall problem at the same time. It also prevents a T4 server from moving to T1 in 3 weeks because (somehow) they won each week in a simple winner-moves-up system. Servers still need to earn enough glicko points within a tier to move up.

It also eliminates the need for glicko modification like artificially changing the scores for Crystal Desert.

One big advantage this method has over others is that it could be implemented without the developers needing to modify the glicko in any way. A couple lines of code in matchmaking would suffice. It should not take significant development time away from other areas. A big time-to-quality-of-life tradeoff.

What you are describing is basically one up one down, has been discussed a lot in the past and has many problems.

For a better ranking system, I would suggest this one :

http://coveragewars2.com

It’s not, the difference is explained in the original post.

Both of these proposals don’t take into account the complications introduced by server linking, it seems. The fact that “host” servers get all the advantages or disadvantages of the Glicko (or score) earned by their alliance while “guest” servers get nothing (good or bad) will create tumult at the beginning of every relink.

Something something inherited score weighted average? (Weighted by %population in previous link? %population in new link? ???)

Tris Apollumenon.6435:

Both of these proposals don’t take into account the complications introduced by server linking, it seems. The fact that “host” servers get all the advantages or disadvantages of the Glicko (or score) earned by their alliance while “guest” servers get nothing (good or bad) will create tumult at the beginning of every relink.

Something something inherited score weighted average? (Weighted by %population in previous link? %population in new link? ???)

No amount of scoring or matchmaking will solve that problem, though. The same servers are the “main” servers at this point, with the 12 remaining servers having to be linked to someone. The “guest” servers glicko ratings are not used at this point.

I’m not sure this is better than a simple 1U1D. The reason being is that it will take longer to fall or rise to an appropriate tier than 1U1D.

Say a T1 server loses a significant number of people. They may not have the lowest rating in T1 and will take time to lose Glicko. And once they fall to T2, they will still have the highest rating there so will be thrown back up into T1 even if they lose the T2 match.

This would definitely be better than the current matchmaking, so would Gudradain’s suggestion. But I still think neither allows for quick enough movement of servers up and down the rankings. And either not the proper variety, or not enough variety in the case of Gudradain’s.

Hesacon.8735:

No amount of scoring or matchmaking will solve that problem, though. The same servers are the “main” servers at this point, with the 12 remaining servers having to be linked to someone. The “guest” servers glicko ratings are not used at this point.

Matchmaking, no, but ranking, yes. What I mean by this is that if host and guest servers alike “inherit” score/Glicko from the number achieved by their mutual alliance, and there is some system set up to divide/weight those numbers when servers are relinked, that could provide a stabler starting point for any Glicko- or score-based matchmaking system.

One possible method, just to illustrate what I mean: all servers inherit the exact score/Glicko from their previous alliance, score/Glicko of new alliance is calculated as an average of component servers’ scores/Glicko ratings (weighted by relative population size in the new alliance so that a small server has a smaller impact on the new alliance’s starting score/rating than a large server).

Alternatively, a “share of the credit” model: servers inherit a percentage of the score/Glicko from their previous alliance according to what % of the total alliance population they occupied, new alliances’ starting scores/Glicko ratings are the sum of component servers’ inherited portions.

Simple 1u1d wouldn’t be affected at all, of course, but your proposal and Gudradain’s both rely on more nuanced metrics of some kind.

@OP

I took the time to carefully reread your post to understand the difference with your suggestionand 1 UP 1 DOWN and… It has the same problem as 1 UP 1 DOWN : it will break a perfectly balanced match up and push a server up anyway.

You need to keep perfectly balanced match up intact as much as possible because those are the most interesting match up you can be in. Variation when nothing is wrong is not a good thing.

Also, your system introduce a new problem that 1 UP 1 DOWN doesn’t suffer from. For example, let’s take SBI falling down to tier 4. Imagine for a moment that DH utterly destroyed SBI in the match. Which means that SBI rating would go down a lot and DH rating would go up a lot. But, since SBI rating was so much higher than DH at the beginning of the week, even after their loss they would still have a higher rating than DH and be pushed up to suffer in tier 3 for another weeks.

Johje Holan.4607:

This would definitely be better than the current matchmaking, so would Gudradain’s suggestion. But I still think neither allows for quick enough movement of servers up and down the rankings. And either not the proper variety, or not enough variety in the case of Gudradain’s.

I just one to point that the problem with 1 UP 1 DOWN is that introduce too much variability.

Also, the system I describe would do very well if a tier 1 server suffer from a mass exodus for example. It would replace that server to the correct tier as fast as 1 UP 1 DOWN but then it would stop once it finds the right tier.

Gudradain.3892:

@OP

I took the time to carefully reread your post to understand the difference with your suggestionand 1 UP 1 DOWN and… It has the same problem as 1 UP 1 DOWN : it will break a perfectly balanced match up and push a server up anyway.

You need to keep perfectly balanced match up intact as much as possible because those are the most interesting match up you can be in. Variation when nothing is wrong is not a good thing.

Perfectly balanced match-ups are a myth, you’ll never get them across all tiers at the same time. For example T3’s matchup is balanced, but T4’s is not. By locking T3’s because it’s balanced, you keep T4 unbalanced.

What needs to be aimed for is the best chance at a balanced match.

Also, your system introduce a new problem that 1 UP 1 DOWN doesn’t suffer from. For example, let’s take SBI falling down to tier 4. Imagine for a moment that DH utterly destroyed SBI in the match. Which means that SBI rating would go down a lot and DH rating would go up a lot. But, since SBI rating was so much higher than DH at the beginning of the week, even after their loss they would still have a higher rating than DH and be pushed up to suffer in tier 3 for another weeks.

That’s not a problem, that’s intentional. That’s a feature I’d advocate for. In your hypothetical it protects SBI from a bad week. It also protects DH from moving up too fast and suddenly getting trounced in T2/T3.

I don’t think it’s a good idea for a T4 server to hypothetically win T4, then win T3, then win T2 and be in T1 in 3 weeks, which is what a pure one-up-one-down system would allow. It also throws past results and progress out the window and gives reason to stop playing in the case of a blow-out.

It also gives servers incentive to play until reset, which now many matches are settled in the first day or two. By clawing away as much glicko as possible to move up or avoid going down, matches stay interesting longer. This is one of the problems that anet is trying to solve by changing scoring. But this is one of those things were the solution does not have to be complicated to work.

It doesn’t have to be perfect, it just has to be better than what we have now.

Hesacon.8735:

Perfectly balanced match-ups are a myth, you’ll never get them across all tiers at the same time. For example T3’s matchup is balanced, but T4’s is not. By locking T3’s because it’s balanced, you keep T4 unbalanced.

What needs to be aimed for is the best chance at a balanced match.

I designed my system exactly to deal with those situations.

Since tier 4 is unbalanced enough it will force a change in tier 3 even if tier 3 is balanced.

On the other hand, tier 1 and 2 are not unbalanced enough to need a change so the system is not changing them at the moment.

I don’t think this works, for the simple reason that Glicko itself doesn’t work any more.

In NA Tier 4 the SBI combo of servers is due to see their overall glicko decrease despite the fact that they clearly have a much higher score at the moment than the other servers (combined!). This means you can’t even use the weekly change in Glicko to determine matches; not to mention the fact that historical glicko is only relevant to solo servers since the score earned by multiple servers is only attributed to one.

At the time of writing this mos.millenium gives scores of:

• SBI/HD: 425 969
• SF/FC/GoM: 226 624
  *DH/EB: 166 408

with SBI/HD glicko predicted to change by -25.552.

I know the situation has an artificial element because of the recent invisible glicko tweak by the devs but any system where a server combo can win so convincingly and yet have less chance of promotion is broken.