Alternative Ranking League


#1

After some consideration, I want to ask if anyone else is interested in participating in an alternative ranking system.

The basic premise is that the only way to get a rank above 30k is winning games, and that people of different ranks must use mandatory handicaps for valid games.

Obviously, this kind of system would include some manual work on my part, but it seems like an interesting experiment to see if this alternate ranking would eventually drift to a very close replica of current ranking of OGS for the players involved, or something much different.

And it’s yet one more excuse to play more games. That can’t be bad.

This is NOT a feature request or in any way related to the OGS server’s ranking system. None of the rules are particularly set in stone, either. This is really just a rough first draft. Suggestions are very welcome.

1.Ranking System

Everyone starts at 30k. No exceptions.
Upon 4 consecutive wins, a player is promoted.
Upon 4 consecutive losses, a players is demoted, unless at 30k.
A maximum rank past 9d may become necessary, but is undecided.

2.Valid Games

A game must be on a 19x19 board.
A game must have at least 30m main time and a byo-yomi which grants
at least 30 seconds per stone on average.
A game should be completed within 5 hours of absolute time.
A game must have exactly zero komi. If for any reason it does not, it will still be judged as if it has zero komi.
If a player plays against a player who is more than 9 ranks lower, the game will not be valid.
A game should contain none of the following:

  • endless wasted moves to bore the opponent into resigning
  • an egregious scoring mistake
  • resigning a won position or a very unclear position
  • trash talking or otherwise harassing an opponent
    As a guiding principle, minor mistakes will simply be corrected since we’re all human, and the silly little details aren’t worth putting people off participating.

3.Handicap System

If a player plays against an equal rank player, they shall alternate between playing black and white. Since strict alternation is a pain, random assignment of color is fine.
If a player plays against a player one rank lower, they must take white.
If a player plays against a player two ranks lower, they must take white and give two handicap stones. For every additional one rank difference, they must give one additional handicap stone.


#2

so if there’s no komi how do we judge ties?


#3

With a tie, neither side gets a win. Any winning streak is broken.

Also, neither side gets a loss. Any losing streak is broken.


#4

I doubt it will be possible to collect enough players for any sort of separate league to function. Look at OSR. It’s super popular, but even their leagues aren’t that populated.


#5

You said (and I paraphrase) that you didn’t really care about the specifics, so I’ll offer my impression of the idea and describe the possible ‘result space’.

Restrictions:

  • League = opt-in = selection bias
  • 30m+x*30s time limit = few games, rank changes very slowly
  • Handicap system not based on established Glicko, but league rank = players will avoid to play equally glicko-rated players with a lower league rank = fewer games
  • Rank starts at lower bound = informational value of losses is greatly diminished
  • Rank = sum(number of 4 consecutive wins) - sum(number of 4 consecutive losses)
    – this means you only start to somewhat reliably distinguish between players 4+ stones apart but in the most extreme cases (players playing in different player-sub-pools), you will not distinguish between players who are 8+ stones apart
    – you will probably have to “restart” the consecutive-win counter after each n=4 wins, otherwise you get an N-3 rank increase (i.e. 7 consecutive wins = 4 rank-ups) due to the moving window

Consider these win/loss strings:

Win(1)-Loss(0) String Winrate Max Rank Adjustment
00010001000100010001 0.25 0
11101110111011101110 0.75 0
00000000000000001111 0.20 -3
11101110111001111111 0.80 +1
00001110000111000011 0.40 -3
11110000111100001111 0.60 +1
11110001111000111100 0.60 +3

Of course these are some of the most extreme cases, but the general idea is clear: don’t throw away information. If you treat WWWL as identical to LLLW, that’s just a lot of wasted effort on the players’ part.

From my point of view, a ranking system should serve the players. If you make it artificially hard to advance, all you’ll see is various exploitative strategies.


Edit: Undid changes to the ‘max adjustment’ column to stay consistent. I treat the ranks as the set of non-negative integers (0,1,2,3,…) as it both expresses the hard lower limit 0 and unambiguously represents our common understanding of increments (+) as progress.


#6

This is NOT a feature request or in any way related to the OGS server’s ranking system.

To clarify - where would the ranks be kept then, and who/what would update them ?

Everyone starts at 30k. No exceptions.

That is fair, but is also going to be hilarious when a new strong player tries to join and all the current 20k players have to give him/her 10 stones.

Also, existing players may have a disincentive to play against the new (involuntarily sandbagging) player. This could make it really hard for the new player to rank up to a point where they wouldn’t be sandbagging anymore.

(smurph) From my point of view, a ranking system should serve the players.

I think that’s right - it would be difficult for a ranking system to make it past the fun-thought-experiment stage otherwise.


#7

My understanding of the motivation for this ranking system would be to encourage handicap play, serving the players by

  1. Providing a stone-equals-kyu rank to compare themselves to others.

There is an assertion that this is a useful thing to have, which I read in chat, but haven’t appreciated why yet.

  1. Causing more handicap play taking place at OGS, helping to calibrate our actual ranking system to 1 stone equals 1 kyu, which we all implicitly think is the case. When actually there is no reason why it would be because we don’t play enough handicap for it to be calibrated that way.

The assumption being that this serves the players by making our ranking system meet our expectations.

It’s an ironic assumption in a way, because since we don’t play handicap much, why do we even care if 1 stone really equals 1 kyu :slight_smile:


#8

If the point was to compare handicap performance to non-handicap performance, the experimental setup would be rather straightforward: either analyze handicap tournaments (granted, that would be an analytical setup) or create a handicap league:

  • N participants play 2(n-1) free live games against their league peers (depending on the number of participants you may have to create divisions)
  • Each season, everyone plays 2 games per peer, perhaps first with proper handicap based on Glicko and the second time ± a stone, based on win/loss in the previous match
  • Compare actual results to results predicted by ratings

Depending on the number and variety of participants, divisions could simply be rank divisions populated by 1 player per rank, ensuring that each game will be a handicap game.