Proposal: Redesign small board komi and handicap

This proposal is three-fold:

  1. Adopt new stone/komi formulas for small boards.
  2. Include komi and board size in ratings evaluations after game result.
  3. Expose “handicap rank difference” in the small board challenge interface.

But first, some background.

0. Background: Description of current system

The existing komi system for small boards on OGS uses the “old Japanese” recommendations for komi (see Sensei’s library) regardless of ruleset. It has some flaws:

  • For even games, komi should match 19x19 and account for area vs. territory.
  • For handicap games, most sources on Sensei’s Library (see above link) suggest that the “old Japanese” handicap recommendations favour black. Wikipedia’s Handicapping in Go page indicates broad support for scaling factors of 2.5-3x on 13x13 and 6x for 9x9.
  • BUG: the OGS komi calculation for small boards confuses handicap stones with rank difference.
  • BUG: the OGS ratings adjustment for small boards confuses handicap stones with rank difference.

Note that these bugs are both quite long-standing (I checked back to 2014). Due to the nature of the bugs, the three parts of this proposal are somewhat intertwined.

BUG: Small board komi is calculated incorrectly on OGS

OGS has a long-standing bug where the komi is calculated incorrectly, caused by the double-meaning of the word “handicap” as both “number of stones” and “difference in rank”.

For 13x13 games, the bug results in:

rank stones old Japanese komi OGS komi
0 0 5.5 5.5
1 1 0.5 0.5
2 1 -4.5 0.5
3 2 5.5 -4.5
4 2 0.5 -4.5
5 2 -4.5 -4.5
6 3 5.5 5.5
7 3 0.5 5.5
8 3 -4.5 5.5
9 4 5.5 0.5
10 4 0.5 0.5
11 4 -4.5 0.5
12 5 5.5 -4.5
13 5 0.5 -4.5

For 9x9 games, the bug results in:

rank stones old Japanese komi OGS komi
0 0 5.5 5.5
1 1 3.5 3.5
2 1 0.5 3.5
3 1 -2.5 3.5
4 1 -5.5 3.5
5 2 3.5 0.5
6 2 0.5 0.5
7 2 -2.5 0.5
8 2 -5.5 0.5
9 3 3.5 -2.5
10 3 0.5 -2.5
11 3 -2.5 -2.5
12 3 -5.5 -2.5
13 4 3.5 -5.5

BUG: Ratings adjustment too big after small board handicap games

OGS has a long-standing bug where the ratings adjustment is…

  • too big if white loses and
  • too small if white wins

… after a small board handicap game. This is caused by the double-meaning of the word “handicap” as both “number of stones” and “difference in rank”.

The current OGS rating system only looks at the number of handicap stones. On small boards, you should also look at board size and komi.

Reference: Handicap and komi on 19x19

(This section is here for context.)

19x19 is the standard board size in all the rulesets, and is the basis for the handicap system. A difference in rank is intended to match the fair number of stones in a handicap game. For more details, see this forum post on rules.

Komi is added in an “even” game to compensate white for moving second. A fair value is worth half a move at the start of the game.

  • Territory rulesets set the komi to 6.5, establishing a territorial value of ~12 for each handicap stone, and adding 0.5 so that white wins if the score would otherwise be tied.
  • Most area rulesets set the komi to 7.5, adding +1 komi for the area value of the stone.
    • Officially, Ing rules set a komi of 8, but black wins drawn games, so the effective komi should still be 7.5.
  • New Zealand rules allow draws and set the komi to 7.

In a “handicap” game, no komi is given.

  • Most area rulesets give bonus points to white to compensate for the area value of the handicap stones.
  • White wins draws (komi is effectively 0.5), except for New Zealand rules, where draws stand.

To summarize:

19x19: komi on OGS:

rank stones territory area NZ
0 0 6.5 7.5 7
1+ 1+ 0.5 0.5 0

19x19: effective komi on OGS (including the handicap bonus):

rank stones territory area AGA NZ
0 0 6.5 7.5 7.5 7
1 1 0.5 1.5 0.5 0
2 2 0.5 2.5 1.5 0
3 3 0.5 3.5 2.5 0
4 4 0.5 4.5 3.5 0
5 5 0.5 5.5 4.5 0
6 6 0.5 6.5 5.5 0
7 7 0.5 7.5 6.5 0
8 8 0.5 8.5 7.5 0
9 9 0.5 9.5 8.5 0

Note: AGA rules use area scoring in way designed to match territory scoring precisely, resulting in a handicap bonus that is off-by-one.

1. Adopt new stone/komi formulas for small boards

Proposal: normalize the komi for even games on small boards to match 19x19. This matches modern recommendations. For example, see:

Proposal: add a handicap stone every 3 ranks in 13x13 games, and every 6 ranks in 9x9 games. This is a compromise based on the various tables and data here:

Handicap and komi on 13x13

The data suggest that the correct scaling factor for ranks / handicaps on 13x13 is between 2.5x and 4x.

Proposal for OGS: adopt a scaling factor of 3x for 13x13.

Why? 3x is in the correct range (broad support for 2.5x to 3x) and conveniently divides evenly into 12 (the territorial value of a free move).

In practice, this means subtracting 4 from the komi for each rank difference, and adding a stone each 3 ranks.

13x13: stones and komi on OGS:

rank stones territory area NZ
0 0 6.5 7.5 7
1 1 2.5 3.5 3
2 1 -1.5 -0.5 -1
3 1 -5.5 -4.5 -5
4 2 2.5 3.5 3
5 2 -1.5 -0.5 -1
6 2 -5.5 -4.5 -5
7 3 2.5 3.5 3
8 3 -1.5 -0.5 -1
9 3 -5.5 -4.5 -5
10 4 2.5 3.5 3
11 4 -1.5 -0.5 -1
12 4 -5.5 -4.5 -5
13 5 2.5 3.5 3

13x13: stones and effective komi on OGS (including the handicap bonus):

rank stones territory area AGA NZ
0 0 6.5 7.5 7.5 7
1 1 2.5 4.5 3.5 3
2 1 -1.5 0.5 -0.5 -1
3 1 -5.5 -3.5 -4.5 -5
4 2 2.5 5.5 4.5 3
5 2 -1.5 1.5 0.5 -1
6 2 -5.5 -2.5 -3.5 -5
7 3 2.5 6.5 5.5 3
8 3 -1.5 2.5 1.5 -1
9 3 -5.5 -1.5 -2.5 -5
10 4 2.5 7.5 6.5 3
11 4 -1.5 3.5 2.5 -1
12 4 -5.5 -0.5 -1.5 -5
13 5 2.5 8.5 7.5 3

Note: AGA rules match territory scoring, just like in 19x19.

Handicap and komi on 9x9

The data suggest that the correct scaling factor for ranks / handicaps on 9x9 is between 4x and 7x.

Proposal for OGS: adopt a scaling factor of 6x for 9x9.

Why? 6x is in the correct range (and has broad support) and conveniently divides evenly into 12 (the territorial value of a free move).

This means subtracting 2 from the komi for each rank difference.

9x9: stones and komi on OGS:

rank stones territory area NZ
0 0 6.5 7.5 7
1 1 4.5 5.5 5
2 1 2.5 3.5 3
3 1 0.5 1.5 1
4 1 -1.5 -0.5 -1
5 1 -3.5 -2.5 -3
6 1 -5.5 -4.5 -5
7 2 4.5 5.5 5
8 2 2.5 3.5 3
9 2 0.5 1.5 1
10 2 -1.5 -0.5 -1
11 2 -3.5 -2.5 -3
12 2 -5.5 -4.5 -5
13 3 4.5 5.5 5

9x9: stones and effective komi on OGS (including the handicap bonus):

rank stones territory area AGA NZ
0 0 6.5 7.5 7.5 7
1 1 4.5 6.5 5.5 5
2 1 2.5 4.5 3.5 3
3 1 0.5 2.5 1.5 1
4 1 -1.5 0.5 -0.5 -1
5 1 -3.5 -1.5 -2.5 -3
6 1 -5.5 -3.5 -4.5 -5
7 2 4.5 7.5 6.5 5
8 2 2.5 5.5 4.5 3
9 2 0.5 3.5 2.5 1
10 2 -1.5 1.5 0.5 -1
11 2 -3.5 -0.5 -1.5 -3
12 2 -5.5 -2.5 -3.5 -5
13 3 4.5 8.5 7.5 5

Note: AGA rules match territory scoring, just like in 19x19.

2. Include komi and board size in ratings evaluations after game result

Proposal for OGS: update ratings system to include komi and board size information when evaluating a ranked handicap game. This will fix the existing bug in the ratings adjustments after a small board handicap game result.

The ratings system needs to compute black’s effective rating, incorporating the handicap, to compare against white’s rating. By adding the board size and komi, we can get the right result for small boards as well.

(NOT in this proposal: allow custom komi in ranked games. However, that would be reasonable to consider as a follow-up.)

3. Expose “handicap rank difference” in the small board challenge interface

Proposal for OGS: allow users to specify a handicap rank difference in the challenge interface for small boards. Currently they can only specify the number of handicap stones, and that’s not enough to determine the correct komi.

Proposed “Handicap” section in the Challenge interface

For all board sizes:

  • Says “handicap” at the top
  • Rank difference: drop down with Auto, Custom, None, 1/2/3/…
  • Handicap stones: 0/1/2/3/…
    • Greyed out, showing computed stones, unless Rank difference is “Custom”.
  • Komi: Auto, None, Custom
    • Custom is greyed out unless unranked
    • Either:
      • greyed out, showing computed komi, or
      • shows custom komi-picker

For 19x19:

  • Rank difference: Auto=default, Custom is greyed out
  • Handicap stones: always computed

For 13x13 and 9x9:

  • Rank difference: Auto=default, Custom is greyed out unless “unranked”
  • Handicap stones: computed unless “Custom”

For other board sizes (always unranked):

  • Rank difference: greyed out, Custom selected
  • Handicap stones: always drop-down

Future extension: rank difference of Auto+1, Auto-1, Auto-2, Auto-3

NOT part of this proposal, but the above interface also leaves room for implementing Auto+1 (increase handicap by 1), Auto-1 (decrease handicap by 1), etc., by putting those options in the “rank difference” drop down. These are common options in live tournaments (see also the “Handicap settings” section on page 45 of OpenGotha’s documentation). See also the feature proposal I posted on the forums before discovering the bugs with small boards.

17 Likes

This sounds like a great direction.

Would it be at all possible to get some data logged somewhere about the auto-handicap setting and rank difference that was used to create each game? Right now it’s almost impossible to check how it works since that information is lost or obscured after the game starts.

3 Likes

Interesting and very detailed proposition, thank you for the work on this !

One question about the 3rd section :

Would this remove the possibility of playing customised handicap games ranked, which users currently have?

For example, would it become impossible to play a 19x19 as a no komi game, unless the ranks were at a difference of 1?

Is there a particular reason to prevent this on all board sizes?

2 Likes

No, it shouldn’t change that (unless I have a bug in my description…). Rank difference has these options:

  • Rank difference: drop down with Auto, Custom, None, 1/2/3/…

Only “Custom” would be greyed out for ranked games. So for your example, you’d set “rank difference” to “1”, which would set no komi.

EDIT: effectively, for 19x19 ranked games, the “rank difference” drop down takes the place of the old “handicap” drop down, and gets used the same way.

4 Likes

That’s a good idea. I think it should be possible…

1 Like

Ahh okay, somehow I read the Custom option as being equivalent/necessary to use the 1/2/3 selector. :sweat_smile:
I see now ^^

I’m all in favor of 9x9 komi depending on the lesser player’s rating. 6.5 or even more may be fair for two 5dan players, but is not fair for a 15kyu player.

What makes you think like this? With no komi 15k or 5d same difficulty coming for white. So the bonus to start is estimated at 6.5 so we give these. Seems fair and simple

1 Like

Agreed. 6.5 for territory rulesets, and 7.0 for area rulesets

1 Like

I’m in favour of those scaling factors.

I use a scaling factor of 6 for 9x9 in my childrens club since 2012.

I used a scaling factor of 2 for 13x13 from 2012 until 2021, but it seemed a bit too easy for black relative to the handicaps on 9x9 and 19x19, so I increased that factor to 3 in 2021 and that seems to work better (better balance across different board sizes).

2 Likes

I’ve been playing 9x9 games for over a year here, and I always find it easier to win when white, because the komi compensates me for being a weaker player. Ideally, I should mostly play black, but I find that when I choose to play as white I win more often. This is especially true in the BadukPop app, where the app chooses my handicap and komi randomly. One of the problems that us 10kyu players have is that it is hard to maintain motivation when we lose more games than we win. I wonder if my statistics reflect this perception (I’ve never been able to understand the four-color statistics functions). Is there another playing site that is better for beginner players? I have been a beginner since learning Go from Bruce Wilcox in 1975 or so, and not for lack of trying to learn. But that’s another story, and off-topic here.

For small boards (bots and humans) you can try Goquest.

10k is not what i call beginner anyway, aren’t you thinking to try the big board?

2 Likes

Playing style could play a role here (and I think playing style is mostly independent of playing strength).

Black needs to play actively to convert their initial sente into a sufficient point excess over the course of the game to pay komi (this goes for all board sizes).
So an active/aggressive player may feel more comfortable with black (bossing their opponent around a bit), while a calm/defensive player may feel more comfortable with white (collecting points at every opportunity).

7 Likes