It seems well known that a method for rank inflation is to play users around your real strength or weaker with high handicaps in sudden death blitz. Nine dan accounts have been made this way, as well as more modestly ranked accounts but inflated nonetheless.

It seems like a lot is done to maintain the integrity of KGS ranks including monitoring sandbaggers and the controversial escaper policy. However, blitz games are rated and can be abused to trick the rating system.

Isn't this a problem with sudden death games, rather than blitz games per se?

Sudden death and go don't work well together. No idea why kgs supports sudden death, but not useful time systems such as Fischer/Bonus or Bronstein/Delay timing.

_________________
Sanity is for the weak.

Jts, I say blitz in general because people can do something like 1m main time with 1x10s or 1x8s byo-yomi. Of course, you can be more specific, and perhaps people would like to discuss what types of blitz should be rated.

For example, maybe 5xT byo-yomi where T>10s should be minimum for rated game?

I'm not answering because my answer might be misinterpreted. I think it might be okay for blitz games to be rated if there separate ratings for blitz and slow games.

It might be hard to get people to agree on the cutoffs, though.

Go Teaching Ladder recommends 20 minutes per side for review

The AGA has these guidelines for rated games, which shows that the definition is hard because many time systems have to be taken into account:



(Source: http://www.usgo.org/ratings/RatingQuals.html)

If we don't rate blitz games and choose too stringent a requirement for rated games, then there are players who would complain because they wouldn't be able to get rated games at all! So there is a tradeoff between keeping online users happy and solving the occasional "fake 9d" and related problems mentioned.

On KGS, the majority of games are blitz if you use either the AGA or GTL guidelines, (and so are almost all games on IGS) so it's impractical to say they should all be unrated.

For this reason, when I hear an online rating, I usually assume it's a blitz rating unless there is good reason to believe otherwise.

If the OP is referring to extermely short time limits (which might show up as Ultra-Blitz on KGS) than that's a separate discussion. Having a rating that's based mainly on one's degree of netlag seems silly to me.

Sudden death is the only time limit that can provide any kind of guarantee that a game will be finished by a certain time. With other time systems, the actual length of the game in moves affects the amount of time they take up. Traditional Japanese byoyomi, in particular, can drag out a lot if both players use all of their time. A 240-move game at 5x30 Japanese byoyomi could take close to 2 hours at the extreme end even before accounting for main time. In most cases it will be finished in a hour, but you never know...

IMHO, the reason sudden death is good and the reason it is bad are one and the same. It just depends on the priorities of the players (or TDs)

Japanese byoyomi, the most popular setting on KGS, is hard to budget because it aggressively spills. The systems you mention are either non-spilling or spill less in practice. I'd like to see one of them on KGS. I'd prefer something non-spilling to provide the maximum constrast to the existing options.

Chess keeps track of lightning, blitz, and standard ratings... They are considered different skills entirely.

Why don't we do the same?

_________________
Tactics yes, Tact no...

Good question. Hmm. If only there were some kind of centralized authority. Some sort of single, despotic arbiter to set these boundaries...someone wise.

Off topic, but at the U.S. Congress this year, IGS Pandanet had a presentation advertizing their new Java client, GoPanda. I wasn't there, but I heard they were trying to attract new users. Having multiple clients and portable ones is a start, but IMHO, a good way to get new users would be to implement some of the common feature requests that aren't being addressed by other servers, like time-category rating classes or alternate timing systems.

Excellent point. Why don't we?

Well, wms frequents this forum, I wonder if we can lure him here to lurk this thread...

_________________
Tactics yes, Tact no...

Don't wonder. I sent him a PM with a link to this thread

... but on KGS, you can! Just make separate blitz and main accounts and you're good

_________________
My name is Gijs, from Utrecht, NL.

When in doubt, play the most aggressive move

Blitz games must be ranked, because blitz games are popular. Removing ranking for blitz would just kick a large proportion of KGS users in the face, it isn't going to happen. I don't personally think it's even a particular problem - the perception seems to be that strength varies more widely as time limits grow shorter, due to the greater probability of silly mistakes, but I don't see how this prevents the ranking system from working as it should. Does it even matter if someone looks 9d but isn't if they never play any games because they'll soon lose it?

Separate ratings might have various advantages, but I think it also has disadvantages. How would these ratings be displayed? Would player names now be listed as 'kgsplayer[4k][3k][1k]' to show ranks at three different levels? Or perhaps 'kgsplayer[l:5k,b:1k]'? I can't think of a way to do it that doesn't add meaningless clutter. If we just display a single rank, we run into exactly the problems that are already being discussed, so the exercise is not useful here.

Finally, none of the different rank ideas stop people from creating new accounts until they get lucky in a few blitz games and raise that rank high. So whilst rating accuracy might be increased at the cost of complexity, I don't think you'd actually solve the suggested problem.

It is useful, if that displayed rank would be the standard rank. Certainly a true 9d can easily attain that rank in standard games if he wants to? If we are saying that some people _require_ blitz games do maintain the rating, then the whole idea of labeling players with a single rank is flawed to begin with. But if we say that ranks are a good approximation of the player's overall understanding of the game, then standard games are certainly a better gauge of that.

If somebody only has a blitz rating, it could be displayed with a question mark in the meantime.

Of course in a game, it would simply show the rank that matches the time settings of the game that is played. And the player information screen can display ranks and ranking graphs for all time settings separately.



It will at least be a lot more time consuming, if standard games are required for the displayed rank. Though I don't really see an issue with somebody making an account with a few games to get lucky and then never touch it again. That's just silliness.

I am inclined to say that if someone can maintain 9d via blitz games, they are 9d, and not a problem in the system. Why is it a problem if someone consistently performs such that their rank is 9d?

Giving blitz ratings question marks still seems to kick blitzers in the face, just less hard (but of course this is more my opinion). I don't really understand the perceived problem with the ranking system here. Why am I not a real 2d if I only play blitz games?

There are a few minor ways to game the system, such as making lots of blitz accounts until one gets lucky and reaches a high level then using it to play slow games against strong players, but when you start playing the slow games the account will quickly drop back down anyway and nobody is really disadvantaged.



I thought this was the problem people were talking about, and would agree that its time consuming nature makes it unimportant to consider.

I was basing this assumption on earlier statements such as 'It seems well known that a method for rank inflation is to play users around your real strength or weaker with high handicaps in sudden death blitz. Nine dan accounts have been made this way, as well as more modestly ranked accounts but inflated nonetheless.'.

This seems to me to not be a problem, for basically the reasons I said above:
- The population of people able to do this reliably seems very small.
- The population of people able to do this unreliably seems unimportant because it's time consuming (as you say)
- There is no obvious problem to me with people having accounts a couple of ranks higher than their real rank. Partly because the number of such people will be small, and partly because I can't really see a way for them to gain any advantage from it.
- If the user mixes slow and fast games, the problem goes away because the slow games (according to general wisdom) won't go so well for them and they will move back to a more reasonable rank. If they don't mix slow and fast games, they (and similar players) are already working on a 2-level ranking system as they don't interact with non-blitz players, and so the problem seems already solved.

Perhaps I have misunderstood the problem, but I still don't see it.

I would agree that there are improvements that could be made to the ranking system, but I don't personally think any of them (certainly not those here) would give an improvement of greater value than the complexity cost.

If it's not broke, don't fix it...

And what's the problem if someone is 5d at blitz, 3d at slow games and is rated 5d if he only plays blitz?

This is just for personal use, though. The complaint seems to be one regarding a kind of rank faking, intentional or not. Maybe if it were a little easier to see a user's most frequently used time settings at a glance without clicking on their games one by one it would be easier distinguish the blitzers from the slow game players. It wouldn't prevent people from becoming 9d using time gimmicks but at least such users would be easier to spot.

I think your assumption that users with inflated ranks who play mixed games (standard and blitz) will soon drop down to a normal rank is wrong. A user can play in the way described to attain an inflated rank then play a couple slow games and (probably) lose then go back to the SD to balance those losses. He may not advance any more ranks, but he is certainly capable of maintaining an inflated rank relatively easily.



Think of it as reverse sandbagging. Sandbagging is a practice forbidden by KGS policy. Also, the problem is not rated blitz, per se, but a certain type of blitz that increases randomness in the result. So having a threshold for a reasonable amount of thinking time for a rated game seems the easiest solution.

To add some perspective, maybe only 9d's on turn-based servers are 'true' 9d's. Not a good argument.

Concerning the poll: I find it interesting that of the eleven people who voted that blitz games should be rated, eight think "SD inflation" does not have to be dealt with. Of course, this may just mean that they think it is not a problem not that they use SD inflation. I wish I added more voting options.