Guild Wars 2

I understand that it calculates a match as each server fighting 2 separate matches against the opposing servers, but the math behind it is a bit over my head so I can’t run through this myself and figure out why we are seeing the outcomes we see and how to best play the system.

I do know that they are changing it, and I’m looking forward to that. But I’d like to understand how it’s working now because some of the results I see don’t add up common sense-wise.

For example:

10. Sea of Sorrows 251 392 1683 +22.650
11. Stormbluff Isle 141 180 1586 -24.035
12. Crystal Desert 157 175 1575 + 1.810

13. Ehmry Bay 262 284 1482 + 7.753
14. Borlis Pass 161 446 1365 -12.196
15. Anvil Rock 173 038 1357 + 4.592

In the first set, SoS is about 100 rating points ahead of SBI and CD, they are beating 2nd place by 110k and 3rd by ~90k.

In the 2nd set, EBay is about 120 rating points ahead of BP and AR and is beating 2nd by 100k and 3rd by 90k.

Overall, the differences in score and ratings aren’t that drastically different, they are in the same ballpark. Yet SoS is netting 22 points for their victory while EBay is only gaining 7. SoS’s victory margin isn’t 3 times better than EBays is, nor are their competition 3 times tougher based on ratings.

So where is the discrepancy? Is the 20 extra rank points EBay has on it’s competitors the big factor? Or is it the extra 10k lead SoS has on it’s #2 competitor that amplifies the gain?

Essentially, what I want to know is in it’s current form, what does glicko give points for? How do we play this system? If it’s going to be another month before it’s replaced, I don’t want to just wait around for that to happen to see a new matchup pairing… I want to try to change it now. So what do we have to do? Does standing on the 2nd place server and creating an astronomical point difference net more points than beating both by about an equal amount? Short of putting up 400k and sending both servers home with their underwear pulled up over their heads, what outcome or set of outcomes does the lead server the most good?

You don’t.

Glicko play you.

Fair to say point gaps are about the same in each tier right? EB is 120 rank points ahead of 2nd. SoS is 100 rank points ahead of 2nd. So it’s fair to say that both servers are 120 rank points above each 2nd place server, which is why EB is getting little points while SoS is getting a decent amount. If the matches were the exact same, both EB and SoS would probably be around 120-130 rank points of 2nd after a few weeks. Of course the real problem is how does one tier even relate to another tier? EB has been in that same matchup for a couple months now, how do we know they won’t destroy tier 4 now? Or that BP and AR won’t be destroyed in T6? Glicko can only give ratings based on the already decided ranks of the servers playing, and since he same servers always fight, it’s nearly impossible to significantly change ranking points enough to actually move tiers.

Khayoss.2019:

Essentially, what I want to know is in it’s current form, what does glicko give points for? How do we play this system?

1. Glicko gives points for your server achieving a higher score than is predicted given the relative glicko ratings of your server compared to the servers you are matched up against.

2. You play the system by getting the highest score possible. If one server has a much higher rating then it could be beneficial to try increase your score relative to theirs, but the predicted behavior of the third server makes this difficult to achieve.

So how do we know what the predicted score of our server vs our two opponents would be? And does marginally beating both predictions equal a larger shift than blowing one out of the water and possibly not beating the other? There has to be a way to know what the watermarks are going into the match and how beating each watermark relates to point gain…

Glicko does not give you points for winning a match.

Glicko give you points if you perform better than expected and it takes points from you if you perform worser than expected.

The expectation is computed from the Glicko ratings at start of the match.

If you have a higher Glicko score than your opponent at the beginning of the match you are expected to win by a corresponding (don’t know the exact parameters) score quotient. If you do exactly this you do not get nor loose points. If you reach a better quotient you gain points, if you reach only a worser quotient than you loose points.

The “watermark” of a match is the average Glicko-Score of a Match.
And this average does (nearly) not change due to a match.
What one server gains the others have to loose (and vice versa).
So the average is (mostly) the same before and after the match.

If you are above average you dominate at least one server of the match, the stronger you dominate (or get dominated) the further you move away from the average.
If you dominate stronger than last week (with the same servers) you gain points, if you dominate less you loose.

If a server of your match goes up and a another drops down your match looses average Glicko-Score (as the up-going has more points than the downcoming). If a server of your match goes down and another comes up your match gain average Glicko-Score.

So whenwever there is a tier-switch of servers Glicko-points are moved from the lower tier to the upper tier. This is the reason why the matches stagnated now after several switches within the 8 month it is now in use.

To get up you need to dominate your match much more to go up than some month ago. And if you finally succeed you increase the distance between the tiers, such that the next switch happens only after even stronger domination.

Ok, so this is good. This is getting closer to an answer. Do the best you, can, gain the most points possible, take those as givens.

If going into a match a server knew that they had to beat server 1 by x and server 2 by y just to break even, that gives them their starting point. Now the server has to establish what its priorities are beyond that initial goal.

If the goal is 50,000 over #1 and 45,000 over #2 to break even, once those are realized, what nets more gain, going for:

100,000 over #1 and 50,000 over #2
or
75,000 over #1 and 75,000 over #2

Or is there no difference, meaning that the rate of point gain is a static 1 to 1 type relationship where each x number of points over a predicted outcome directly equals 1 rating point gain regardless of which opponent it came from?

If you score is above score average your end-glicko-rating will be above glicko-average.

For the exact number you have to use the formuals, I don’t think it is a linear function.

Dayra.7405:

I don’t think it is a linear function.

I don’t either, that’s what I’m hoping to find in an answer somewhere here… someone smarter than me that maybe knows the math and can tell me “Hey, when you reach that base level, go smack server #2 more than #1 because pound for pound that lead is going to be worth more.”

There is also some inertia included as well. (see EU-T1 at millenium.org).
Vizu does not loose all (but a large portion of) it’s rating advantage, just because its score is below average.

The best way of playing Glicko is a regular look at http://mos.millenium.org/matchups#EU and “get a feeling for it”

Fix this glicko rating system. It’s broken and makes matches boring as hell. Who wants to continuously play the same servers every week that shouldn’t be there? Let the fluctuations of points move greater. If a blowout happens, it happens for a reason and it means that the server is not in its proper place. Make the movement of points greater. This joke of a system is making WvW boring and repetitive with no end in sight. Not to mention its poor rewarding system. Kind of explains why some people are feeling that GW2 is easier to get bored when compared to the original GW.

It’s not Glicko that is broken. Glicko is perfect for ranking based on (random) matches. The mistake ANet made is to determine the next match based on the ranking.

If you would separate Ranking and Match-Making, e.g.
-. looser down, winner up, or
- random matches, or
- probabilistic matches, or
- a liga-system like proposed in https://forum-en.gw2archive.eu/forum/pvp/wuvwuv/Suggestion-WvW-2-0, or
- …
to determine the matches, Glicko would work perfect for the rating.
Just the feedback loop of rating based on matches and matches based on rating was a big mistake.