Some ideas and suggestions and general thoughts:

1. This has been suggested before, but I would like to see how many wins and losses and the percentage of games a player wins on their ladder profile. It is possible for people to calculate this anyway, so it's not like displaying this information is presenting any information that is hidden or otherwise unknowable. I really like statistics and comparing and contrasting them and evaluating them, and not having this information readily available without having to go through and manually calculate it is time consuming and causes me to not want to do the data comparisons I would otherwise like to do. Things I want to see: Win percentage and total wins/losses/games played. Things I want to look at: the win percentages of the top 20 (or more) ladder players and how this correlates to their placement in the ladder and whether or not total games played affects it.
Also, if anyone doesn't want people to be able to see this information on the games they have played, there can be an option for them to select to hide this information (much like how the actual profiles work), which would be no different than how it currently works.

2. Some sort of basic statistics displayed next to the ladder on the home page that show fun statistics in addition to seeing the ladder rankings. I mean things like "fastest movers" showing the people who have gained the most ranks on the ladder in the last week or "longest winning streak" showing the player who currently has the most consecutive wins. There are plenty of other stats that can be shown here as well, but basically this would just be things to make the ladder look more appealing and generate more interest in it.

3. Sorting game history on a player's ladder profile. It would be nice to have links in the headings that toggle ascending/descending sorts on things like opponents ratings, date, and game result (whether the game was a win or loss). This would allow someone to look at a player's ladder history and separate their wins from their losses, or look at their results in order of opponents rating to see how they fared against the highest or lowest rated opponents they face. Sorting by date allows a player to see how well they are faring over time (whether they are winning more lately as they improve, etc.). Sorting by date in descending order could be the default (which is how it is now).

4. Show previous matchups and results between two players on the game page. By the game page, I mean the page that provides a link to the game, says who got first pick and who won the game and provides links to each players normal and ladder profiles. This would be a simple section below the current information that shows prior matchups and who won them as well as links to those previous games. This would be helpful information if you're on someone's profile page and you want to see how they have performed against some other player in the ladder. Rather than having to crawl through every game they've played and manually calculate, you could simply click on one of the games and see the information readily.

Something that I have previously lobbied to Randy for is a win/loss statistic that would be something like this:

Win/loss versus Top 10:
Win/loss versus 10-25:
Win/loss versus 26-50:
Win/loss versus 51-100:

etc.

The above breakdown is just an example and was chosen quite arbitrarily.


It would be very similar to what you see in NCAA basketball "tournament resumes" where the wins and losses for that are often based on RPI, but we would use ladder ranking.

I think this directly correlates with your first point above.

Thoughts on the ELO system and Pairing Algorithm:

**Pairing Algorithm**: Changing it to only pair you against players who are within 25% of your current spot is working. People are starting to settle out into spots that seem reasonable, and there are far less "big swings" in the ladder due to pairing anomalies. I am no longer seeing players catapult to the top of the ratings "by fluke" or without having to play tough opponents and beat them to get there. I think this was an almost strictly positive change to the ladder.

I know some people dislike it because they are going to play similar opponents frequently without getting to experience playing everyone on the ladder, but I couldn't disagree more with that sentiment. Playing everyone means that there are going to be large number of "no win" games between people with huge rating differences. By that, I mean games where a much higher ranked player is going to get no points for winning it (and is risking a massive point drop if they lose) and the lower ranked player could be easily frustrated because they are playing a game they have a very low chance of winning. If the higher ranked player wins, then both players gain almost nothing of value from the game (the higher ranked player could lose points simply for having played the game using the current rating system!). If the lower ranked player wins, that one game is going to artificially inflate his rating and deflate the higher ranked player's rating.

**ELO System**: I still have complaints with the system. I know people just recently voted to keep this system in place (I voted for the "other" option, personally), but I want to show some of what I would consider flaws. First of all, losses affect your rank much more than wins do. I think that I have to win between 10-15 games in a row to surpass Doushibag on the ladder, who is currently only 23 points ahead of me (2058 to 2035). My reasoning behind this is that I have won 4 straight games since beating him (against the 4th, 5th, 14th, 17th ranked players), and each game has caused me to gain between 2-3 points in rating relative to his rating. The reason these gains are so small is that every time I win, he gains rating points as well because he's played me 3 times, and since he has played a significantly smaller number of games than I have, my rating significantly affects his rating. His recent win against the 18th ranked player created more separation between our ratings than my wins against the 4th and 5th ranked player combined.

What does this all mean? It means that there is really no way for me to earn my way to the top of the ladder by winning games, even against top 10 players. The only way I can surpass him at this point (without grinding about 10 straight wins against top 10 players without losing any games, which is a near impossible feat) is for him to lose to someone else. The best way for him to remain at the top of the ladder is to simply not play games. This doesn't create a healthy ladder environment. Once you get to the top 10 in the ladder, the best way to stay there is to simply not play. It's discouraging for players who are trying to raise their rank in the ladder because they are reliant not on their own ability to win games, but on the players above them to lose games. If you don't get a chance to play the people above you, then you are at the mercy of other players to beat them. There's very little feeling that you control your own destiny. So much of it is up to chance and up to the honor of the players above you to play games at a reasonable rate and not stall games.

My best chance for being number one on the ladder is to simply not play any more games until Doushibag loses to someone. Each game I play increases the chance that I widen the gap between me and him on the ladder if I lose. By winning, I've seen that I'm gaining almost no ground on him, 2-3 points per win, even though all my recent wins have come against top 20 players.

Does this matter? Only 25% of the players in the top 20 of the ladder are playing more than 2 games at a time. That means that 75% of the top 20 ladder players are playing 2 or less games at a time, which is the minimum number of possible games. I have mine set to play 5 games at a time and I only have 3 games right now and I'm not getting paired up with new games. Considering I can be paired against anyone in the top 20, this means that there are no free games anywhere between anyone in the top 20. People are not playing games, and they are playing what games they have slowly. I have not gotten a new game in almost 24 hours! That means nobody in the top 20 has finished a game in the last 24 hours, or if they have they are being instantly paired up against someone else who is also unable to get games.

This is not healthy for the growth of the ladder. People should have to defend their spots in the top 10-20 of the ladder. There should be punishments for people sitting on a rating and stalling their games, or conversely, there should be a benefit for people who are winning games. If you're in the top 10 of the ladder and someone below you wins 5 games in a row, they should pass you on the ladder. If you're not playing any games, your rating should start to drop as people who are playing games and winning are passing you by. If you want to move back up the ladder, your option is to play games and win them. People should be able to challenge people above them on the ladder. If you're in 4th place and the 3 people above you are sitting on their rankings and you can't possibly pass them without them losing, then you should have an option to challenge them and make them earn their right to retain that top spot by beating you.

I really think that the current system is hurting the health of the ladder. I find it personally discouraging that someone who has played 1/4 the games I have can sit ahead of me, play the minimum number of games and that I have no chance of passing him without him losing. My only hope is to be lucky enough to get paired against him so I can beat him or to be lucky enough to have someone else beat him for me. There is no incentive to win games, only to not lose games. My current settings of playing 5 games at a time is actively harming my chances of being the number 1 ranked player. A ladder system that rewards not playing games is not a healthy system.

Sorry I am so long winded. I know people are going to just tl;dr my last post, but I think it raises a pretty reasonable complaint, so I do hope people take the time to read it and respond.

To demonstrate what I mean, I just ran the program a few times to see what would be necessary for me to surpass Doushi, who is only 23 points ahead of me right now.

I would have to beat 3rd place Fizzer 6 consecutive times in a row without any other games finishing on the ladder to tie Doushibag in rating at 2068. I would gain 33 points and he would gain 10 points from me beating Fizzer 6 times in a row. I gain slightly less than 4 points on Doushi per win against the 3rd place player.

If I lose once to Fizzer, I drop 9 points and Doushi drops 2 points. I lose 7 points in difference by losing to the 3rd place player. Looking strictly at these two stats, I have more to lose than gain by playing any games against anyone who isn't Doushi. Since beating Fizzer 6 straight times is both impossible based on the ladder pairing algorithm and would also be near impossible based on the pure chances I have of winning that many consecutive games against a skilled player, there is no incentive for me to play games unless I'm playing against Doushi himself.

On the other hand, if Doushibag loses to any player on the ladder--doesn't matter who it is--and I don't play any more games, I pass him on the ladder.

Is there any incentive for me to play ladder games? I can't pass Doushi by winning games, unless I win at minimum 6+ games in a row against top 10 players and Doushi doesn't win any games during that period. I can only pass him if he loses. I just demonstrated this by running the bayeselo simulating different results. The answer is no, there is no incentive for me to play games. I should reduce my games to 1 at a time and stall all games I'm currently in if I want to achieve first place. During that period of time my fate is entirely in the hands of another player in the ladder beating Doushibag, something I have no control over at all. That is how I can achieve the number 1 spot on the ladder. What fun is that? The best way to be number 1 is to not play? That's lame. That suggests to me we need a different system.

The only reason I'm playing games right now is because I enjoy playing games. Each game I play actively hurts my chance of achieving the number 1 rank. The rating system is supposed to encourage people to play games by providing incentives for winning. This rating system encourages people to not play games, because there isn't a reward for winning, there is simply a punishment for losing.

I'd like to second #2 in the original post above. Also it would point out if the ladder is becoming static.

Yeah, as my long winded post in the other thread on the voting for rating system showed I too think this system is flawed. Not enough people recognized that and the votes suggested to Fizzer that everything was good enough and not worth all the extra effort to change things. Perhaps when another better system becomes more readily available without a large time investment he'll take another look at things. But until that changes we have to suffer with the current system it seems.

So the solution is to make one of the better systems more accessible to implement. If you can do that I'm sure he'll have another look at changing it from the current system and then problem solved.

On another note you will get to see #1 again... it's only a matter of time before I slip and lose and it could easily be sooner rather than later (since I'm not delaying anymore too). Interestingly when I plugged in a potential win/loss for my last two games (NuckLuck & Siggy) somehow beating the higher level person was more critical to maintaining my score than the lower person. Which made no sense to me. This system is quite odd in its valuations. And by odd I mean it seems pretty stupid to me. And in general more games equals less movement in this system. So the more games anyone plays the less they'll move. It's trying to solve where exactly it thinks someone belongs, not flowing around with how you do. If you improve the system won't take account of it until early losses are off the books 2-3 months later.

Don't worry, there are definitely people working behind the scenes on getting a suitable implementation of an improved rating system set up to present to Fizzer :)

|> First of all, losses affect your rank much more than wins do.

I think what you're noticing is only true for the players with amazing records, such as those near the top of the ladder. Doushibag only has two losses, so one more loss increases his losses by 50% which, of course, has a large impact.

The inverse would be true for someone near the bottom of the ladder - wins affect your rank much more than losses.

|> I think that I have to win between 10-15 games in a row to surpass Doushibag on the ladder, who is currently only 23 points ahead of me.

Look at this from the perspective of the rating system. The rating system considers all games within the last 3 months equal - newer games aren't considered any more relevant than older ones. Doushibag has defeated you twice, whereas you have only defeated him once. Further, Doushibag has a 88% win ratio, whereas you have an 83% win ratio. Even if you win your next 15 games, that only brings you to a 86% win ratio which is still lower than Doushibag's.

Of course, win ratios are not everything, given that who the wins are against is important. But given that Doushibag also has a winning record versus you, these stats don't really feel wrong to me.

Now, if you had a winning record against Doushibag it would be a different story. Then, even with a lower winning ratio, I would feel like you should be on top. If you only had an even record against Doushibag, then it's not as clear - his better ratio makes it feel like perhaps he should still be higher. But since you have a losing record against him, the results feel correct to me.

Look at it the other way - if someone were ranked above you who you had beaten twice **and** had a lower winning ratio than you, wouldn't you feel something was wrong?

|> If you're in the top 10 of the ladder and someone below you wins 5 games in a row, they should pass you on the ladder. If you're not playing any games, your rating should start to drop as people who are playing games and winning are passing you by.

It sounds like what you want is for newer games to "count more" than older games, or for players who play more games to rank higher than players who play fewer games even if they're slightly less skilled. Those are the only ways I see to modify the system to make your scenarios actually happen, as today it considers all games within the last 3 months equal and doesn't give benefits to players with more activity.

|> The only way I can surpass him at this point is for him to lose to someone else.

The problem with Doushibag's rank on the ladder is that he was allowed to get to the #1 position with a relatively low number of games. I [covered this](http://blog.warlight.net/index.php/2011/03/ladder-players-be-sure-to-vote/) on the blog, but it was basically a loophole that used to exist with the matchmaking algorithm that allowed this to be possible. Doushibag observed how the system worked and found a way to rocket to the top. What you're experiencing now is really my fault for not doing more to prevent it from being possible in the first place, and for that I apologize.

The good news is that I believe the problem has largely been fixed. Players joining the ladder now have to work their way up the ladder from the bottom, which makes it so you can't rocket to the top by virtue of having a small sample size.

Now, when I fixed the problem, I realized that this would only fix it in the long-term. Doushibag's existing rating would take some time to even out. It wouldn't really be fair to take action against him since he didn't break any rules, he just played the ladder system to it's maximum efficiency. He's been drawing it out, but it will happen eventually.

Also relevant, my policy on slow play in the ladder is this: I'm fine with players taking their time when they still have a chance to win, but if you've clearly lost, please surrender. This means you can't run away from the opponent with an income of 5, taking 2.9 days for each run. If you see someone doing this in the ladder, shoot me an e-mail. I had a talk with Doushibag and his recent surrenders happened due to this policy. His future games are also being monitored.

Given that it's fixed in the long-term, it's hard to justify spending time changing the way the ladder system works just to get over the short-term hump. Look at it this way, if it takes a week of work to change the ladder system, but the problem would have fixed itself in three weeks, it's hard to justify spending that week of time on something that will only benefit players for two weeks when I could spend that time on a feature that will live on the site forever, benefiting players for much longer than two weeks. This is obviously a simplification, but I hope the general idea makes sense.

Thanks for the response. You make a lot of good points. I don't agree with some of it, although I'm not sure whether or not it is worth responding to the bulk of it. I guess I'll pick and choose a few things to address.

|> Look at this from the perspective of the rating system. The rating system considers all games within the last 3 months equal - newer games aren't considered any more relevant than older ones. Doushibag has defeated you twice, whereas you have only defeated him once. Further, Doushibag has a 88% win ratio, whereas you have an 83% win ratio. Even if you win your next 15 games, that only brings you to a 86% win ratio which is still lower than Doushibag's.

|> Of course, win ratios are not everything, given that who the wins are against is important. But given that Doushibag also has a winning record versus you, these stats don't really feel wrong to me.

To some extent, I agree, but not entirely. Yes, he has won 66% of our matchups and has a higher overall win percentage, but I don't think those two stats tell the whole story. The more games you play, the more your win ratio is going to even out or settle to what it should truly be. Example: Player A is 99-1 with a loss to player B who is 10-0 with a win against Player A. Who should be ranked first? I don't think there is enough information to make a clear judgment in that situation, but I would strongly lean toward player A being in first. He has a worse win ratio and a loss to player B, but his 99-1 record is simply more impressive (assuming quality opponents, etc.). I guess this is a subjective call, but I think most people would agree A's rating is more impressive.

Looked at another way: If you take my first 19 games on the ladder, I had a 94% win rate. My first 25 games I had the same win record as Doushi currently has now (22-3), an 88% win rate. It is by playing more games against stronger opponents that my record deteriorated from that point to the 83% that it is now. In other words, I've basically been punished for playing games while being high ranked, whereas the best option would have been to beat lower ranked players until I got high ranked and then stop playing.

I know I'm making this about me (mainly because I know me and my stats and how I've done on the ladder), but I think this is true for anyone who has played a large number of games. Bostonfred, if you take his first 25 games has a 21-4 record. That 84% win ratio is higher than my current 83% win ratio. It's likely if he stopped playing after 25 games that he would be 2nd place on the ladder right now, instead of the 8th he is in now. His rating is punished for playing more.

|> It sounds like what you want is for newer games to "count more" than older games, or for players who play more games to rank higher than players who play fewer games even if they're slightly less skilled. Those are the only ways I see to modify the system to make your scenarios actually happen, as today it considers all games within the last 3 months equal and doesn't give benefits to players with more activity.

Actually, I don't want either of these things. What I'm mostly trying to do is point out what I consider a serious flaw with the system. The flaw is that when you are highly ranked there is no value or incentive to play games. You are only hurting yourself by doing so. The path to achieve that high rank in the first place is going to include games against easier opponents as you climb in rank. Therefore for the majority of players, your win/loss and win percentage are going to be higher than what they should be since it factors in those games. Also, the less games you play the more variance you are going to see in ratings. I don't want a system that rewards people who play more games, I want a system that doesn't directly punish them for doing so!

|> The problem with Doushibag's rank on the ladder is that he was allowed to get to the #1 position with a relatively low number of games. I covered this on the blog, but it was basically a loophole that used to exist with the matchmaking algorithm that allowed this to be possible. Doushibag observed how the system worked and found a way to rocket to the top. What you're experiencing now is really my fault for not doing more to prevent it from being possible in the first place, and for that I apologize.

I don't think you need to apologize for this at all. Things happen, it's impossible to foresee how things are going to pan out. I know my posts here may sound ungrateful, but I'm actually really thankful for all the things you have done for the ladder system, including the polls and willingness to make changes based on suggestions. It's why I'm even writing these posts in the first place, because I see a serious flaw with the system and I know that by vetting it there's a chance it will be fixed and changed to a better system. It's my frustration with the current system and my hope for a better one that is fueling my posts here.

I don't think I'm articulating very well what I see as a problem. The main thing is that in a ELO system where player ranks are calculated relative to other players, and not based on their individual wins and losses, the only games that matter to achieving the number one rank are games against players higher ranked than you. The only thing you need to do is have the system view you as a stronger player relative to them. So if you're in 3rd place on the ladder, there is no value in playing any games against anyone who isn't 2nd or 1st place on the ladder. You are already ranked higher than the players 4th and lower on the ladder and playing them only increases the chance for them to surpass you by beating you. Since you can't control who you play, and since you have a very high chance of being paired against players lower than you on the ladder because there are significantly more of them in the pairing range, your best option is to not play games and hope that someone else beats the top two players for you.

This system **favors** people who play less games over people who play more games. What I'm not calling for is a system that favors people who play more games, but rather a system that doesn't directly **punish** people who do! More games means more chances to lose to people ranked lower than you, which means more chances for the system to consider them to be a higher relative skill. More games doesn't necessarily mean more games against the higher ranked players either. If they are playing the minimum number of games and playing slowly, it could be months before you get a game against them if at all. What does that mean? It means that there is a non-trivial chance that you don't have a chance of passing someone ahead of you on the ladder without relying on the skills of other random people to beat them and that every game you play directly hurts your chances of being high on the ladder.

Again, 75% of the top 20 players are playing the minimum number of possible games. I see this as a problem. Every game I play right now that isn't against Doushibag is a no-win proposition for me. I stand nothing to gain from it and a lot to lose from it. My best strategy is to reduce my games to 1, and make that statistic "80% of the top 20 players are playing the minimum number of games."

Doesn't anyone else see this as a serious problem? A system that provides no incentive for high ranked players to play games and all the incentive for them to sit on their rating is not a healthy system!

Here are the two main things I think the solution should be: Challenges. If I'm in 2nd place, let me challenge the first place player to a game that he can't refuse. Make him earn his 1st place ranking by having to fend off players trying to wrest it from him. Currently, the 2nd place player has to hope that the first place player's games against the 13th and 16th place player that have been ongoing for a month and a half end up giving him a loss. That puts the fate of the 2nd place to become first place player 100% in the hands of other random people and not at all in his own hands.

Secondly, a more traditional ELO system that gives you points for each game you play. This DOESN'T mean you have to play a lot of games! You can still be number 1 by playing a small number of games if you are winning them! It just means you can't sit there indefinitely by stalling or not playing games and that other people aren't punished if they do play games. This is a more "Active" system that encourages people to play games at high rankings, rather than encouraging them to stall games, which the current system does. If you're in first place, and the 2nd place person rattles off 7 straight wins against top 10 players and passes you, you are encouraged to finish stalled games and win a game or two to move back ahead of them. This encourages active gameplay where people actively vie for the top spot. Again, this doesn't mean you have to play a lot of games. If you're winning a high percentage, you can be the top with a low number of games, maybe even lower than what is currently possible. However, you can't stall once you get there to remain in that spot forever, especially if people are able to challenge you.

Finally, I don't think players slow playing speeds and extremely low number of current games was factored into the equation when the ladder was developed. Consider this. Doushibag plays 1 game at a time. He moves every 2 days. The average game is 15 moves. He finishes 1 game a month, or 3 games over the 3 month period that is used to calculate ratings. Where is the ladder going to place him when it only looks at a 3-0 rating over the last 3 months? That is certainly something to think about. What happens to players who achieve a high rank and then drop to sloth speed on making moves? Doushi could conceivably achieve a 2250 rating again by simply winning some minimum number of games every month (and stalling or hoping to avoid losses based on matchmaking). We have no clue how the ladder is going to factor this in.

I'm curious how something works as there are two possible ways for it to work and I'm not sure which it is. But if I fall to below 30 counted (games in last 3 months) then I will have to start playing 2+ games again, or no? Additionally if I fall below 10 counted games would that not take me off of the rankings until I got back up over 10 again. Meaning if I got to 30+ games and then played slowly for 2-3 months and only finished 5 or whatever games in the last 3 months that I'd fall out of only being able to play 1 and fall out of being ranked. Meaning I'd have to play back up to 10 to get back into the rankings again and play back up to 30 to get back to the point of only being able to play only 1 again. Is that how it works or are rating-irrelevant games counted in the 10/30 limits?

Another point I would like to reiterate that I think I made in my other long ladder post was that Warlight has a distinct problem that other games don't have when it comes to ratings. That's that games take longer and thus ratings will always in comparison be a bit deficient as not as many games are played. As they're almost by nature going to have to deal with a smaller sampling size to rate people with. In other games they can easily complete in 30-45 minutes per game and finishing 1+ game a day is fairly reasonable. That's just not how Warlight goes as it's a slower game. Thus even a better system will still have to deal with a smaller sampling size. Adjusting down for rating confidence could level this out somewhat to encourage more games, but that's only a partial solution and could end up punishing people who don't play enough and underrate them.

I do not know if any of these ideas could be easily implemented, or if they have been thought of before, but I will share a couple of them.

What if the player ranked first had to have 5 games going?

What if the player ranked first had to make moves every 2 days?

What if we used a capped banking boot timer? The timer would start with 72 hours grace, and every time a player made a move they would add another 24 hours to the timer, but the timer could never have more than 72 hours on it.

What if only the last 30 games of an account were used in the determination of rank? Games would still time out after 90 days.

Ahh, Doushi, why'd you drop off the ladder? That saddens me.

|> What if the player ranked first had to have 5 games going?

I'm not sure if this is a good idea. A lot of people don't want a system where someone is required to play a lot of games. The current system is very well suited for not playing a lot of games, but provides no incentive to play more than the minimum number of games, and actively punishes top ranked players who do.

|> What if the player ranked first had to make moves every 2 days?

Not sure I agree with this either, because some players don't play on weekends, and those players would be prevented from being the number 1 ranked player based solely on that. I also don't like anything that pertains only to the number 1 ranked player, but that doesn't address the system as a whole. Why stop at number 1? Why not the top 10? Why not the top 20? It's hard to draw the line and say "only this subset of players on the ladder have to play by a different set of rules than some other subset."

|> What if we used a capped banking boot timer? The timer would start with 72 hours grace, and every time a player made a move they would add another 24 hours to the timer, but the timer could never have more than 72 hours on it.

I really like this idea. I think Blue Precision (maybe someone else?) suggested an idea similar to this before, but I think this is a really good idea. People who can move every day during the weekdays will always have 72 hours on it, except for weekends, and they can build back up to 72 hours again during the following week. This would shorten the overall time games take to finish (by forcing people who are around but are stalling games anyway) to play faster.

|> What if only the last 30 games of an account were used in the determination of rank? Games would still time out after 90 days.

That's an interesting idea. Not sure exactly what I think about it right now. It would certainly get rid of the issue of someone only having 3-4 games in the last 3 months to base a rating off of or having less than the provisional 10 games. It could result in wild rating swings as people complete games and old ones fall off. Might be worth trying out.

Anyone want to take bets on how long I can sit in first place by dropping to one game at a time and moving once every 2 days in it? I think I can sustain over 2 months without anyone having a reasonable shot at passing me. I think the only people who could possibly do it are new players who haven't lost yet.

I don't want to do it, but I will if it is necessary to show how fatally flawed the current system is.

|>Ahh, Doushi, why'd you drop off the ladder?

Why did I join it?