You mean instead of using decimals for a higher resolution?
It would be a more coarse resolution when using only 3 steps instead of 10 steps to subdivide a rank range.
1 Like
Itās fitting to people mind (weak middle strong)
1 Like
Iām actually in favour of that. 2k-, 2k, 2k+ as the rank display, with graph and granularity provided by the Glicko numbers which sidesteps the whole decimal debate.
3 Likes
Like this I suppose (which seems sort of a mix of the features from systems A and B to me, as in the rank unit value of all labels matches the basic rank value over the whole range, and the basic rank label matches the middle of the range):
Sidestepping the decimal question is fine for me, but there were some objections to dropping decimals. And I suppose that discussing various alternatives to decimals could lead to wandering completely off-topic.
5 Likes
With the voting results up to now, I think we can draw some conclusions about the decimal rank display system on OGS:
- System A (OP proposal) is not considered to be much clearer than system C (the current system), though we should note that system C may have the incumbentās advantage.
- Of the options presented, system B is considered the least confusing way of displaying decimal ranks, although system B is not perfect either.
There is some concern that changing the decimal rank display system just like that will result in numerous complaints from confused/dissatisfied users. So if OGS will go ahead with changing its decimal rank display system, it may be better to include that change in the next larger rating/ranking system update, whenever that will take place.
2 Likes
Note that if we switch to system B, half of the players will be promoted (e.g. everyone between 1.0k and 1.5k who are displayed as [2k] will be displayed as [1k]), so if we want anchor OGS 1d between AGA 1d and EGF 1d, we need to demote everyone by one-half rank (or change the formula ln(rating/525) x 23.15).
I suppose it depends on what OGS meant when they anchored OGS 1d between EGF 1d and AGA 1d.
Did they mean an average OGS 1d, or weak or strong? And with which EGF 1d and AGA 1d were they comparing that OGS 1d anchor? EGF 1.0d means āaverageā 1d, while AGA 1.0d means āweakā 1d.
If the OGS 1d anchor is the average 1d, and they compared it with average EGF 1d and average AGA 1d, I donāt think anything needs to change. Only the decimal presentation of average OGS 1d will change from 1.5d to 1.0d.
On every server/go association, people have a definite (integer) rank. Your rank on OGS is what is displayed next to your username. Your EGF is what is displayed in the EGD, e.g.
So I suppose the goal is that the average OGS 1d should be between the average EGF 1d and the average AGA 1d. Although I doubt itās possible to be that precise.
I think itās the other way around. The [2k] presentation will not change (the rating range will stay the same). Only the decimal rank of all players will change by 0.5 ranks.
All OGS [1d] will stay OGS [1d] when the decimal ranks display is changed, whatever the new system (A-D).
4 Likes
Oh you mean: those whose decimal rank is 1.4k under the current system will be 1.9k under system B, and those who are currently 1.6k will be 2.1k? Looks good.
3 Likes
Exactly 
In all of the proposals, your first example will remain a slightly stronger [2k] and your second example will remain a slightly weaker [2k].
The question is which decimal rank presentation conveys the most clearly that both players are [2k] and also that your first player is a bit stronger than your second player. Iād say that this is what the OP proposal and the polls are about.
Itās more a matter of User Interface Design than mathematics.
1 Like
Uhh, I thought we were discussing shifting the decimal ranks under the integer ranks, so noone would be promoted or demoted from it?
1 Like
I am a bit surprised that system A got so little traction.
Perhaps I did a bad job pointing out its potential advantage.
I mean, once youād be used to it, it would require minimal mental effort to understand that 3.1k means [3k] close to [2k] and 3.9k means [3k] close to [4k].
I can definitely understand why the AGA uses this system.
4 Likes
You understood correctly.
1 Like
People are probably just confused because they know that System B expressed in a mixed base with decimal for the integer part and balanced ternary for the mantissa would be optimal, and arenāt sure which system is closest to that /hjk
(Note that the - NULL + suggestion is one digit of balanced ternary in effect, so Iām not the only one thinking this)
Yes, Iām very suprised that people are getting so confused by system A. I think theyāre just overthinking it and trying to find superfluous mathematical properties (for that we have the Elo/Glicko system; whereas the kyu/dan system is intrinsically quirky and honestly, beyond saving).
In practice though you just canāt get it wrong. Once you get used to it (that is, you accept that itās just stupidly simple and that thereās nothing to think about beyond 3.x is 3k), itās very clean and convenient
Also people tend to be more used to rounding and means than to flooring and medians, even though the latter are more relevant in some cases.
1 Like
Actually, system B (rounding) has serious issues that arenāt immediately obvious and that seem to not have been understood or at least havenāt been mentioned
- It suffers from essentially the same issues as system A at the kyu/dan boundary (gap in values and ratings on either side that differ only in the k/d mark, e.g. 0.5k vs 0.5d; same as 1.0k and 1.0d in system A)
- The 0.5k and 0.5d ratings are each only 0.05 points wide, instead of 0.10 for other decimal ratings
- Without further adjustments, both 3k and 4k players may have a 3.5 rating (if their precise rating is for instance 3.52 or 3.48) so that the integer rating cannot be deduced from the decimal rating
Fixing #2 and #3 will make the whole thing much less elegant. For instance if to fix #3 we round to decimal then to integer, then the 1k rank becomes 0.95 points wide. Possibly #2 and #3 may be fixed concurrently by making the gap between kyu and dan 0.9 points instead of 1.0 points, so that the hole in the ratings is [-0.45, 0.45[. At this point though (double rounding and custom intervals) weāre in the shenanigans realm
Anyway, the point is that system B is a lot less intuitive and functional that what it is portrayed as. Arguably it has confused even its proponents
1 Like
I donāt think thatās right. Unless you saying that 2k- (2.67k) would round down to 3k?
The people who like 2k, 2k-, 2k+ all being 2k should also appreciate 2.0k, 2.5k, 2.9k all being 2k.
3 Likes
Issues 2 and 3 can be avoided in the following way. Let
R = 30 - ln(rating/525) x 23.15.
Currently,
- 2k rank corresponds to R between 1 and 2
- 1k corresponds to R between 0 and 1
- 1d rank corresponds to R between -1 and 0.
System B could consist in displaying
- If R>0: display the decimal rank floor(10R+5)/10 kyu
- If R ā¤ 0: display the decimal rank floor(-10R+5)/10 dan.
1 Like
To be honest I still donāt really understand what the proposals are when it comes to the decimals 
All I can say is that system B visually made sense.
I think thatās probably the problem, you have to get used to it, and it doesnāt make sense in a mathematical sense that 3.1 is close to 2.
Iām not sure if that can be reconciled really.
Iām not sure any of these issues are actually serious and arenāt just completely cosmetic
2 Likes