Please show me any player on the ladder who loses less than 11.1% of their games.

not me

8/32=11.1%?

I don't think there is any player who can sustain an 8-1 win ratio in the long run. I temporarily have 18-2 record, but it will change for the worse. Possibly top 5 could sustain 80%+ win ratio?

Right. Which is why it's a bit ridiculous for somebody to say that top players lose less than me when I've won 8 of my last 9 games.

24-8 exactly 75% which is also my winrate in non ladder games. I guess thats about my real winrate. Though I'm getting better so maybe I can improve that to 80% long term.

But you do understand that getting better rating means playing against better players?

billy jack of hearts is being a dick to us

But you do understand that getting better rating means playing against better players?

I'm not sure what this has to do with what I said?

Also I'm not so sure that getting a better rating means playing against better players. Lots of top players in the ladder play lots of weak opponents. The algorithm needs some work. People should be playing those ranked close to what they are. If they win, they should be playing people with slightly better rankings, and if they lose, slightly worse. Until it figures out how they stack up relative to everybody else.

Right now a lot of the rating has to do with how well you play against people who are a lot worse than you, relative to how the other top players play against people a lot worse than them - with some of it having to do with how well the top players play against each other, but not nearly enough imo. It should be ONLY about that. What does it prove if HHH wins against people ranked 50th or lower 96% of the time while unknownsoldier wins against them 93% of the time. That means HHH is better? That's dumb. It should be how they do against each other.

And before you get all butthurt again - by saying the algorithm needs work, I'm not implying that your rating is wrong or that mine is wrong. You're obviously very good and one of the best players. That doesn't mean the algorithm is perfect though.

The algorithm is described in warlight wiki. You can read all about it. Again, you are not the first person who is saying "the algorithm/ladders need some work". My question is what is the alternative you propose? What type of algorithm would you use? ELO and bayesian ELO are not something invented just now for warlight, it is a system that stood the test of time and is very reliable from statistical point of view. Answer my question specifically and then we can talk or, even better, Fizzer could potentially implement it.

As I said, I would have people be matched up against those close to their rating.

So if you are rated 1900 and are winning, you would play those rated 1900-2000. If you kept winning, you would get a higher rating and start playing players rated 2000-2100. If you kept winning, you would....etc.

But if you were losing against players rated 1900-2000, you would get a lower rating and play people rated 1800-1900. If you started winning against them, you'd get tougher opponents and a better rating - if you lost, you'd go down again, until it figured out where you belonged.

I know it does something sortof like this now, but there are A LOT of games played against people who are rated nothing close to you. If your rating is 1500 it should never match you against somebody rated 1900 (except maybe if you are very new and it's trying to figure out where ou belong). It's worthless data to figure out the rating of the 1900 player. If he beats you or loses to you, it shouldn't effect his rating, but it does.

Basically I think it should be setup so that basically everybody is winning only about 50% of their games.

Take for example a hypothetical 100% accurate ranking of the following players:

15th
16th
17th
18th
19th

The guy ranked 17th is going to lose > 50% of games to the guys ranked 15th and 16th, but win > 50% against the guys ranked 18th and 19th, and these would be the people he played on a regular basis. So his win % would be about 50%. The same all the way up and down the ladder, except at the very top and the very bottom, I guess. The #1 player, if it never changed, would win > 50% and the worst player, if it never changed, would lose > 50%. Everybody else would be at about exactly 50%. Otherwise they would move up or down until they were.

What you described is exactly what algorithm does. What you forget is that there are <15 people with over 1900 rating and new games are created every couple of hours... Unless you want to force people to play more games that they want at once, what you propose is impossible. Again, please read warlight wiki on the specifics of the algorithm, understanding why each part is essential is key to further conversation. Btw. I was paired with someone below 1700 rating only once (except my first games obviously) and always when another player from top 5 ended a game we were paired together (vide zibik and Oliebol).

Only thing I found on the Wiki about this was:

WarLight ladders use an ELO rating system, similar to what is used in professional Chess tournaments, to rank players by their skill level. Based on the results of your ladder games, you will receive a rating that will be used to give you a rank on the ladder's scoreboard.

Your rating will change based on the results of other player's games, too. For example, if you beat a player who ends up becoming the #1 player, you get the full results of beating the #1 player even if they were not #1 when you beat them.

Players must complete a certain number of ladder games before they are given a rank: 15 in the 1v1 ladder, 10 in the 2v2 ladder, and 5 in the seasonal ladder.

Only members can join ladders, however anyone can view all of the ratings and ladder games on the website.

This doesn't address why it is the way it is or how it got there, or why it isn't or can't be the way I proposed. It just states how it works.

Billy, your solution doesn't take into account the fact that Warlight skill isn't linear; every player has strengths and weaknesses making them susceptible to particular opponents and stronger against others. For example, in your setup, what would you do if 16 beat 17, 17 beat 18, and 18 beat 16? Rankings are only approximations of average skill. The only truly accurate rating system would require an impossible amount of games.

What you described is exactly what algorithm does.

No it doesn't.

This is zibik, ranked #2



Last 10 games he played:

-4th
-31st
-10th
-39th
-22nd
-19th
-35th
-29th
-14th
-38th

Only one in the top 5, two in the top 10, four in the top 20.

50% of his games were with people ranked 29th and below.

Something wrong there.

Another interesting thing there:

He is 3-6 against people ranked 1840 and above. But he is undefeated against everybody else.

So he's very good against people a lot worse than him, but he's really bad against people of a similar skill level. Yet he has a higher rating than all but one of them!

Yeah we have finally understood the real reason why you aren't placed first in the ladder ranking, despite you are the best player.
Warlight is wrong!

---

Actually the funny thing is that even if you will ever be (just to do some unreal hypothesis) the best player, no one will agree with you, mainly for this pathetical thread.
Don't you think you'd just better stop embrassing yourselves?

what would you do if 16 beat 17, 17 beat 18, and 18 beat 16?

It would never be so simple. There would be enough data on how each did with a number of different people to have tiebreakers and such. But yes, that would add a layer of complexity to it.

However my system would avoid things like I just showed. Zibik should not be ranked 2nd if he loses 2/3rds of his games against people in the top 20 - because those are the only people he should be playing, and if he lost 2/3rds of the time against them, he wouldn't even be in the top 20!

Yeah we have finally understood the real reason why you aren't placed first in the ladder ranking, despite you are the best player.
Warlight is wrong!

I've already said this has nothing to do with my rating. I'm not suggesting changes for selfish reasons.