Why is this - if I’m 3.0k ± something, I don’t think it means I’m a weakest 3 kyu, it just means my mean rank is about 3kyu.
If I’m 2.5k ± something, it means sometimes I play like 3kyu sometimes like 2 kyu. Or with a higher deviation adjust appropriately no?
(Ok well it’s more about confidence estimates a bit still that also where I would suspect a system to place a player like that)
I don’t know why this is. With system C, I’m just describing the current system OGS uses.
From your description, I suppose system B makes most sense to you.
System A seems close to how the AGA rating system works (https://senseis.xmp.net/?AGARatingSystem)
The EGF rating system doesn’t actually show decimals, but perhaps it somewhat implies system B to more mathematically inclined people?
I haven’t answered the poll because it asks the wrong question, and it disregards how the system actually works. The system makes perfect sense within its own rules (i.e., the decimals are rounded up across the whole range at the kyu level and down at the dan level). Your system C is how OGS actually works, while A and B are interpretations that people may prefer. Even this is misleading, however, because the discussion since my last post has been divorced from rating points.
Most players (including me) do not know where the benchmarks are for each rank. The purpose of the decimal “ranks” is to give someone an idea of their rating-point progress through the continuum from one rank to another. Nothing more. I think that’s a useful purpose. Arbitrarily altering the decimal would sever its connection with the rating points. People are still free to think of it as they prefer. Optimists can regard 3.0k as the start of their march toward 2.0k. Pessimists can think of 3.0k as the brink of a precipice where they might fall to 4.0k.
If you knew those benchmarks and you saw a rank indicator of 3.0k, would you expect it to match the benchmark for an average 3k, the benchmark for the weakest 3k or the benchmark for the strongest 3k?
This seems to imply that you feel that 3.0k is the weaker threshold of the 3k range, so you’d favour system C (which OGS currently uses). I mean a “start of a march” seems like a longer distance to go than the “brink of a precipe”.
I agree that C is how OGS works currently, as I wrote in my previous post (I wrote that before I read your post that preceded mine). Because of this, I don’t think I can put myself into your hypothetical. OGS could just as well have rounded everything down for kyus and up for dans; however, to do that would alter the benchmarks, for then, 2.9k would be the benchmark for 2k, thereby altering the meaning of the benchmark in terms of rating points. Averages would be even worse, since we could then argue about the meanings of 3.4k versus 3.6k. The current system at least has the logic of following most advancement systems: you complete a set of requirements and you become the higher rank (in go, as measured by rating points).
Well I suppose we’re maybe interpreting what the system is doing differently are we? Is there one absolute reading of it?
Perhaps I should give another example.
@Groin’s current OGS rating is presented as 2k without decimals. The OP proposal wouldn’t change this, so let’s use that as a given.
His rating reveals that he is in fact 0.1 ranks weaker than the middle of the rating range for players that OGS labels as 2k <=> 0.6 ranks weaker than the treshold between 2k and 1k <=> 0.4 ranks stronger than the threshold between 2k and 3k.
What would you expect that his 2k rank looks like when adding a decimal?
I’ve voted for System B. I think “rounding to closest” is most natural/intuitive.
They’re all a bit awkward because there’s hole in the number continuum due to “0” not really existing on this scale. Every one of them looks either reasonable or wrong, depending on how you’re squinting.
Another way to phrase the choice is, where do we want the hole in the continuum? Do we skip/delete just one side of “0”, or both sides, or a little from both sides?
[1k, 1d) (delete both sides of 0 entirely)
[25k, 1k) union [1d, 9d+](6k, 5k]: “5.0k or weaker”[5d, 6d): “5.0d or stronger”[0.5k, 0.5d) (delete half of each side of 0)
[25k, 0.5k) union [0.5d, 9d+](5.5k, 4.5k]: “roughly 5.0k”[4.5d, 5.5d): “roughly 5.0d”[0k, 1d) (status quo; delete the dan side of 0)
[25k, 0k) union [1d, 9d+][5k, 4k): “5.0k or stronger”[5d, 6d): “5.0d or stronger”(Where a parenthesis ( or ) indicates a non-inclusive range and a bracket [ or ] indicates an inclusive range.)
FTR, I liked system “A” before I wrote out the above, but now I think it’s the least intuitive in practice. I don’t like the dissonance between 5k meaning “5.0k or weaker” and 5d meaning “5.0d or stronger”. Thus, I switched my vote to system B. I also think system C (the status quo) is just fine.
I’m also curious what other servers that share fractional ranks skip/delete from the continuum. Does anyone here know?
See
If everyone is displayed as being almost a whole rank weaker overnight then we at least need to plan for the tsunami of “why am I suddenly a stone weaker this morning” and people just generally losing it.
Not necessarily a reason to do or not do something but a surprising number of people get surprisingly agitated about smaller changes to their ranking than this…
Let’s wait the 1 april then (fool day)
Yes, we 100% agree.
Personally I just can’t make sense of system C (other than it being the status-quo), so for me the real choice is between A and B, with A being flooring-based and B being rounding-based. IMO it would be easy to get used to either in practice. Nobody would be surprised if the integer rank changed around the x.5 boundary, but people would also very quickly notice if all 3k users (and 10k, etc) always have 3.x (10.x) decimal ranks.
I’m still on the flooring side. I find it more visually consistent and pleasant (3.x is 3k) and I don’t mind the system going in opposite directions for kyu and dan, because that’s just how that scale works. Once you get under 1.0k, you “fall off” the kyu scale and move to the dan scale, starting at 1.0d.
I think the rounding approach is trying too hard to make the kyu/dan system behave like a proper numerical scale, and this doesn’t work well particularly at the kyu/dan border. 0.x ratings are unsightly, and the jump from 0.5k to 0.5d is awkward, especially as both of these decimal ranks would be half as wide as normal (0.05-rank points wide instead of 0.10, with 0.5k being [29.45, 29.50[ and 0.5d being [29.50, 29.55[).
I’d like to maintain the current system. If it changes when the rating system reset hits, that’s probably fine, but until then it seems like needless confusion
I think each of these systems has its pros and cons. It’s a matter of taste/opinion which one makes most sense to different people. That’s why I created the poll.
The AGA rating system uses decimals like system A does, except AGA uses negative and positive labels instead of kyu and dan labels.
The EGF rating system doesn’t show ranks decimals, but rating graphs show rank lines and labels at the middle of each rank range, implying system B.
KGS doesn’t show rank decimals, but rating graphs show rank lines and labels at the lower threshold of each rank range, implying system C.
So each system has at least one major proponent out there.
In the poll, system A (OP proposal) initially took a lead. Then later system B caught up, while system C (status quo) stayed close behind.
In the mean time, it seems system B took a clear lead:
It seems that system C got at least some votes because it is the status quo, so it might have gottten the least support if the status quo were system A or B.
Indeed. I don’t think this potential issue should be dismissed lightly. System C has been the status quo on OGS for a long time, and even though it may not make the most sense to respondents of the poll, it’s not a bad system.
Yes, shelving the matter until the next rating system update could be an option.
Perhaps this question is worth another poll?
Examples. This player has GoR 2049 and his rank is 1k (the red curve is in the white area but just below the horizontal separation between white and blue)
and that player has GoR 2050 and his rank is 1d
I meant that in EGF rating graphs, the integer rank labels are displayed in the middle of each rank range. Rank ranges are displayed as a colored band above and below the integer rank labels.
The integer player rank matches the integer rank label closest to their current rating (i.e. system B).
KGS on the other hand shows rating graphs like this:
The latest integer player rank of both players is 1k, so the integer rank labels (and lines) are drawn at the bottom of each rank range. The integer player rank does not match the closest integer rank label, but the integer rank label below their current rating (i.e. system C).
Let’s put it another way.
We start from this, where each system agrees on the rank labeling of specific rating ranges.
Then we add rank labels with 1 decimal that is equal to 0.
System A:
System B:
System C:
From there you just interpolate other decimal values, for example ranks with decimal 5:
System A:
System B:
System C:
[Edit:] System D:
I’d like to note that the labeling of OGS rating graphs seems not fully consistent with system C:
For example:
According to system C that OGS currently uses, 3.5d means “average 3d” (and that’s also what it shows in my user profile with [3d]), but the axis label on the right says 4d, implying that 3.5d means “weak 4d” as it would in system B.
Why not (in a hypothetical rating update; I don’t think this should be changed until then if ever) get rid of the gap by labeling the range [2k, 1d] like so?
Then the numbers have a meaning as distance away from shodan, with dan and kyu marking positive and negative numbers by convention?
So in the kyu range that would be equal to system C, but in the dan range it would be different from systems A, B and C.
I suppose that justifies calling it system D:
