Kirby wrote:As I said, I'm just kidding with you. I'm also curious if there exists any admin on KGS that dislikes the KGS escaping system... But I don't want to get into some heated argument about this....*snip for space*
Oh I knew you meant your comment in good fun, as did I (= , my goal was to simply try and get people to discuss objective aspects of what is desirable in a system rather than simply vent because they are angry about escapers. I was actually trying to keep my personal views out of it as much as I could, instead outlining the set of principles I believed were the main focus when implementing this escaper policy. Like I said, this is an old discussion that has been happening for the better part of a decade now. I think most people forget that the current escaper system is actually already a compromise solution from what wms originally had wanted (and if I recall correctly, it was a change that was made reluctantly at that...).
Redundant wrote:I accept that the KGS system works in most cases, However, it might not be the best possible system (I'm assuming that wms wants KGS to be the best server around). I'm wondering how you feel that the policies on other servers are worse then the KGS system in respect to the criteria you mentioned.
I'm not totally familiar with all of the intricacies of other servers escaper systems, to try and do a brief comparison:
Tygem - It was mentioned in another thread you submit your games for adjudication. This would go against criterion 1 as there is a human element now directly involved, it would also increase the effort required by the average user in order to have a game resolved. In theory, this could have an advantage with criteria 3, because (should they be willing to put forth the effort) now all games can be evaluated on a case by case as to what the effect of the various outcomes of the game could potentially be. It's hard to compare the two systems with respect to the 2nd criterion, it would depend upon the specific person making the judgment.
IGS/Oro - As I understand it there's a time limit to log back in (5 minutes?), then you resign. This is fits nicely with 1, it's completely automatic. On the second criterion I feel this falls short however, because the server behavior seems to assume the player is guilty of escaping, and gives them 5 minutes to prove they were not. When in doubt, the server would default to the conclusion that the player is not acting in good faith. I will come back to #3 in a minute. With respect to #4 it's pretty much a wash. While technically on IGS the average player is required to either wait 5 minutes before moving on (because their opponent disconnected), or log back in within 5 minutes, requirements that don't exist on KGS, in practice on KGS they will end up doing this anyway.
It's difficult to gauge the impact with respect to #3, because it turns into something of a signal to noise issue. In general you will have two types of disconnections, someone who is willfully disconnecting due to the board position (i.e. escaping or sandbagging), or someone who gets disconnected (willfully or unwillfully) for a reason unrelated to the board position. Let's assume from a ratings perspective, the former is dominated by escapers and (if left unmitigated) would introduce a systematic bias for select people who engaged in that behavior. Let's also assume that disconnections for reasons not related to board position, the disconnected player had an equal probability of winning or losing that game. With respect to criterion #3, the goal of any automatic escaper system would be to reduce the bias introduced by intentional disconnections while limiting the random error introduced by unintentional disconnections. The IGS and KGS systems take alternative approaches to this. The IGS system's goal is to completely eliminate the systematic bias by making all escaped games resignations, while taking steps to mitigate the random error (instead of all random disconnections resulting in a loss, now only disconnections of greater than 5 minutes result in a loss). KGS on the other hand attempts to virtually eliminate any random error that may be introduced by disconnections (as it would generally take greater than 10 random disconnections in 6 months against opponents who were either unable or unwilling to resume the game), while implementing a system that would minimize the systematic bias introduced by intentionally escaped games (no more than 10 games per user per 6 months, provided you are playing a sufficient number of total games).
We could try and speculate as to the magnitude of these effects, and then decide which is more important to reduce, however I think would be more productive to instead briefly look at the nature of each server's ranking system. On IGS a player is given a rank that's a number that is slightly adjusted each time a game result is entered. On KGS all game results of the previous 6 months are used to generate a probability of each player winning vs. each other player. In short, KGS's ranking system has increased sophistication, however that added layer also means it is potentially more susceptible to bad data points (games have a 6 month memory). Another way to look at it is this - IGS's ratings are made by consistent modifcations to a static value, whereas KGS's rankings are made by repeatedly performing a dynamic measurement.
From this perspective it makes sense that both servers take the approach they do. On IGS, a systematic measurement bias would be more likely to cause a player's rating to continue rising (you would be consistently adding more than you are subtracting), and it could potentially be more harmful to the system. A random error would just mean that the player's rating fluctuates a larger amount, but still averages out around the "correct" value. A systematic bias on KGS however would make it so that a player is consistently ranked a small amount higher than the "correct" value and would not have a cumulative effect. In that vein, for KGS it's more effective to try and limit the random error in order to improve the accuracy of rating system as a whole. Without being able to find the actual magnitude of either of these effects it's tough to say which is doing a better job, or what the best place to put the parameters at to tune each system would be (i.e. Who knows, maybe IGS would have a more accurate ranking system if they used 7.5 minutes instead of 5 and maybe KGS should use 8 games instead of 10...but at that point it's just nitpicking).
