No sub ratings for distinct variations, some may differ quite a lot from rating by name

games I've played which may differ

Einstein vs Backwards capture        requires quite a different game plan

Morelli (sizes)        pieces/space ratio changes quite a lot so maybe more changeable

Havannah (sizes)        maybe not too different

Dots and Boxes (sizes)        would not expect much either

blackhole versions        maybe not much either

Shogi has the same problem.  I am OK at Mini Shogi, but with the big Shogi I often feel lost.

Yeah, +1 for GO. Variants change the game completely, especially the random 100 moves. Separate ratings would be awesome.

100% agree. In go is is even worse for 2 additional reasons:

1. The small boards and the large boards do not have the same number of levels. I try to explain: A is 200 Elo point stronger than B means: A will beat B 75% of the time. From novice to pro, there may be 30 times 200 Elo points in 19x19 and just 15 times 200 Elo points in 9x9. For that reason, matching the two scores simply does not work. It you take the Elo points from 19x19 to 9x9 they are no longer Elo points. (Internally LG uses Elo, then it makes some conversion to kyu/dan, but that does not change the fact that it is a complete misconception.)

2. Computers are near pro level in 9x9 and just a few dan in 19x19. Computers than are near 4 dan in 19x19 are using special hardware and special software not available to normal public. A normal program on a laptop is 1 dan at most in 19x19, but it can be too strong in 9x9 for pretty much everyone here.

For EWN (Einstein würfelt nicht) there is another point to make: playing 1-points match, 3-points match, 5-points match, 7-points match, ... up to 50-points match. The more points the match is, the smaller becomes the luck factor to win vs the skill factor. But for the rating, every match-type has the same weight. So, as YHW pointed out earlier, the current elo rating is not very accurate for EWN players. But I don't mind, I enjoy playing it. The other games are fun too, but for serious playing they require much more thinking time to invest.

Erratic: has anyone run numbers on 9x9 for what you claim? I certainly believe it, but then again, I'd never have believed proper komi on 9x9 could be as high as 7 until I was told it was.

The number of levels in terms of standard deviations can be measured analyzing tournament data. I have read that go has 30-something levels, which is about twice of what chess has. But unfortunately I don't know if someone has analyzed 9x9 go because it is usually not treated as a real game and there is probably not much data. It would be nice if someone can confirm that. For me it just makes sense, but it is not proven by data, afaik.