You and OP are not alone in this.
A survey among Dutch go players (some years ago) showed that 60% was higher educated and 60% was employed in some IT/CS related job, so I think many (Western) go players have some mathematical disposition.
Also there are quite a few OGS forum regulars who are actual mathematicians (even some PhDs).
In one of Robert Heinleinâs books (The Number of the Beast) the term PhD is described in the following fragment:
âOn this campus it swank to assume that everyone holds a doctorate. Even i have one, PhD. Do you know what that stands for?
~ Doesnât everybody? I have a PhD too. Piled Higher and Deeper.â
Many years ago I encountered one of the most sensible thesis statements ever: A person who has a PhD knows everything about nothing.
This is derived from a very old joke definition of specialization, which is âwhen you know more and more about less and less, until eventually you know everything about nothing,â
One interesting way would be to have a hysteresis: coming from 11k you need to reach -10.0 to be awarded 10k and you keep it until you drop to -11.0. That would reduce the rank fluctuations around the rounding borders.
Whatâs wrong with fluctuations?
I like that if my rank drops, getting it back is only one win away, rather than a lengthy road.
Makes me remember how ranks are so huge drama for a lot of players.
I wouldnât mind lower fluctuation. I donât really care about the number much, but a friend recently remarked that I was sandbagging because Iâm 9k now (usually 7k, occasionally 4k)
It seems like the ranking system would be a bit more useful if it recognized that Iâve probably stabilized after some years
I understand that point. Then it would probably be better to slow the actual rating change for players with low variability some more.
Adding hysteresis could make you look like a sandbagger for longer, by letting you stay a 9k longer, even though you barely broke into 9k and immediately started winning again.
I think one issue is that OGS updates rating per-game instead of per-time period. This means that we are finding a solution that works for players who play 5 times a year AND players who play 5 games per day (im somewhere in the middle)
If we could somehow separate the two, it would make sense to have frequent players have less volatility than infrequent players.
I think quite a lot of people prefer that it updates per game.
Maybe psychologically when you lose itâd be better if there was no change, but when people win itâs the opposite, they want to see the points go up 
But it might just be that frequent players simply are more volatile. If you plays 10 game a day, you might perform worse that if you played one good game a day where you had time and concentration and werenât annoyed from the last two or three losses in a row, it might make your wins/loses less predictable.
Perhaps it was hoped that Glicko-2 would somehow handle both of those scenarios well?
Is it that Glicko-2 isnât really capable of that, or is that OGS isnât using Glicko-2 properly?
Youâre right, Glicko-2 does take time into consideration. As far as Iâm aware, OGSâs implementation does not.
On the technical side, I think the reason theyâve done it this way is that G2 expects every player to be processed every period. Thereâs no well-defined period for an online server, and even if we just pick one, OGS would need to set up a regular bulk job for it. For what itâs worth, I assume OGS is operating at a scale where this is possible; Iâm mainly just trying to provide an explanation for why things are the way they are.
I can buy this.
I donât buy that frequent players are inherently more volatile. (Though could be convinced if you have some data to support it)