Yuck. This would be an abuse of the anti-stalling feature, exploiting the fact the current system is poorly implemented as an anti-not-resigning-when-you-are-losing feature, and surely confuse beginners that their games get randomly ended because they suck so bad their opponent can start passing in the middle of the game (which is a bad example for teaching beginners about when you should pass to end the game and score, something beginners already struggle without triple passes in middlegame).
5 Likes
Fascinating effect … which of course makes sense: if the person loses against a weaker person, then we know for sure they are weaker.
I guess “plays only 5k” is an extreme: they can’t keep doing that anyhow, as their rank drops…
4 Likes
How do these compare to the prohibited option of:
- playing against established 25ks
4 Likes
This is a great point! Here is a graph with scenarios that account for this rule:
The 9 stone diff rule helps a lot in comparison to the “5k only” scenario, but a player who plays as high as possible will still remain above 22k* until game 60 or so. That’s pretty bad in comparison to normal matchmaking, which would put the user below this threshold after about 8 games.
*since 25k is the asymptote, I think it is more illuminating to examine when the user crosses a threshold slightly above, such as 22k
3 Likes
Not significantly better than just playing 9 stones down, which is allowed. Both cross the 22k threshold after about 3 games. Keep in mind - in this simulation, we expect the player to win more often against 25k, which won’t give very much signal to the rating system at the start.
Edit: I didn’t account for the 25k rank cap, so “plays 9 stones down” is also technically illegal once the user passes 14k, but since the green and orange lines are virtually identical, I think it’s safe to say this doesn’t matter.
3 Likes
For someone at 25k that would mean playing established 25ks wouldn’t it?
I should clarify, “normal matchmaking” means pairing with established users of equal rank from the system’s point of view. A brand new user looks mid rank (~12k*) to the system, even if their “true” rank is 25k.
For example, here is a simulation of 5 games with the “normal/proper matchmaking” scenario:
| user rank | opponent rank | result |
|---|---|---|
| 11.8k ± 8.4 | 11.8k ± 1.3 | L |
| 15.7k ± 6.8 | 15.7k ± 1.6 | L |
| 18.6k ± 6.3 | 18.6k ± 1.8 | W |
| 16.4k ± 4.8 | 16.4k ± 1.6 | L |
| 18.2k ± 4.6 | 18.2k ± 1.8 | L |
*Is the starting rank actually 1150 glicko (12k)? Or is the rank and rating shown on user profiles adjusted for humble rank?
2 Likes
It’s 1500 rating and 350 deviation. It’s the humble rank of 1500-350=1150 that’s shown initially though, and used for matchmaking, with the automatch at least I believe.
because this is actually possible with the humble rank in automatch, which is much better than
3 Likes
Crud.
Revisions to the previous graphs...
Directionally, the these are the same as the graphs that started at 1150, but things take a bit longer to settle. Note: I still do not take humble rank into account for these, I’ve just updated the initial rank.
The difference between “9 stones up” and “even” is more pronounced. It now takes 200 games to get to 22k (compared to 25 games for the “even” player)
Interestingly, the “9 stones down” player performs slightly better than the “25k only” player now (1 fewer game to get to 22k)
Very true - It looks like humble rank places raw beginners about 3x faster than without it. Very similar to someone who plays 5 stones down, but not as effective as playing 9 stones down.
3 Likes
Is there something amiss with the simulation? 25k should’t be approached asymptotically, as far as I understand it our glicko ratings go all the way down to zero. It’s only our display of them that stops at 25k.
1 Like
25k is the asymptote because the user’s “true rank” is set to 25k. Here is the same simulation for a 5d:
3 Likes
Ah, I could go out on a limb and suggst that a true beginner is rather worse than 25k 
I guess it’d be interesting to see whether it makes a difference to how quickly they hit 25k if they are actually 30k, just while we are messing with it.
I think that to productively continue brainstorming solutions, we need to know if “ask them if they are a beginner” is an option.
I guess these graphs help show why it seriously is needed.
6 Likes
I think also that this gives credence to the effectiveness my “provisional ratings” theory of methodology,
where a chosen rating is applied provisionally and separately from the soon-to-be established rating, starting from 1500 and deviation 350
(although one could argue that maybe the deviation should match the ratings distribution of the current playerbase, or some account of what a “new account distribution” might look like in practice)
but also a second rating with similar (maybe smaller?) deviation is applied until the two have significant enough overlap to be considered established, thus giving an unestablished player a chance to prove their rating among peers, which will likely create some of the fastest convergence to true ratings, and those who are wildly off with their estimation a big flag over them such that if they converge too slowly, their initial assessment was likely wrong.
The downside (or upside) is this might allow a sort of re-initialism by re-flagging a player as provisionally rated at a new rank, thus allowing them to have matches at their intended updated rank after a long time off, but also creating the situation that was often cited for removing the choice of initialism back when the decision was first made – that of moderators being inundated with requests to have their rank manually changed
2 Likes
This could probably be avoided by allowing players to state their level only when they create a new account.
3 Likes
For reference, one more confused beginner:
1 Like
Well, I must confess I was confused as well about what happened. Seems like this guy created an account, went on to play some game, his rating tanked. He then invites a friend who is probably as new to the game as he is and tries to play against him. But the friend is now 12k or something while he is 25k and the default setting is a rated game and they can’t play against one another.
Sure, the app could handle this better, but still, quite puzzling to a beginner why he can’t play against his friend.
7 Likes
This is a recurrent problem i pointed out since years, and there were a new thread again on this opened a few weeks ago, if you are interested on the subject.
Edit: here you are, sorry 
I’ve said this before. I HATE this system. It is too focused on the math instead of the user experience. I don’t give a crap about the formula being butchered and changed out for a new system if it improved the new user’s experience. I much prefer fox. I create an account, pick my rank, and then I get to play without having to waste a bunch of time getting to where I want to be.
New user’s experience should trump the sandbagging gate. Instead there should be a separate system in place for looking for sandbaggers. (Perhaps using AI to red flag account for a mod to review if a certain number of mistakes are used.)
3 Likes
I agree that user experience is vital.
It’s unfair to characterise this as focussed on the maths though: its also focussed on the user experience of us all when the ranking system is messed up by different changes. The challenge is that this kind of experience can only be explained by maths.
I agree that the experience of both new and older players with the current situation is bad with how things are - it’s a continual thorn in the side of older players and a wacky joining experience.
That’s why I keep prodding this conversation, to try to flush out an answer.
5 Likes
this is actually the kind of thing Whole History Ratings was built for, and might be an algorithm worth looking into, assuming it’s not too computationally expensive, which it might very well be.
3 Likes