thanks for the wrap up...!

so what is current system? W-L record, then score? didn't tom change it to be W/L, head-to-head, and finally score?

I've been following this too, so far it sounds like the Conan system is far better than the current and is somewhat explainable to non-nerds  as I'm following it.

Is there a drawback to the Conan system?  Other than the slight possibility of still a pure tie?  (In which there probably isn't a good system is there?  Even manual wouldn't be able to resolve some ties.)

One last question, this only affects when 2+ players have the same number of wins right?  Ie a player with 7 wins and 2 losses wouldn't somehow ended up losing to a player with 6 wins and 3 losses (due to this SOS/SOSOS stuff), right?

One last last question, this would mean tournament scores wouldn't be needed at the individual tournament level at all right?

weathertop wrote:

thanks for the wrap up...!

so what is current system? W-L record, then score? didn't tom change it to be W/L, head-to-head, and finally score?

 

I really don't believe it's been changed.

I hate to throw in a wrench but this is probably the time to put it out there.  In general I think in the RR where let's say two players are tied, whoever won the head-to-head will typically have the greater SODOS number.  But I can imagine the case where the leaders have more than one loss each.  Maybe A lost to B and to the lowest scoring player but B lost to two 2nd place players.  In that case A may end up with a higher (or the same) SODOS.  So do we want any additional checks after SODOS to handle the head-to-head result?  There are a few options:

1. [Wins-->SODOS]  Ignore special H-2-H consideration.  Ties still possible (do whatever we're going to do with ties).
2. [Wins-->SODOS-->H-2-H]  Use H-2-H results as final tiebreaker.  You could still have a tie if both lost in their H-2-H (3 player game, for example).
3. [Wins-->H-2-H-->SODOS]  Apply H-2-H results before SODOS consideration.  Ties still possible.

I think any of the three options could be justified but most players probably wouldn't want to see the H-2-H winner lose.  I think options 2 and 3 are the most fair.  I think 3 sits best with me.

Sounds like 3 is the best to me.  I don't think the H-2-H winner should lose if both have the same number of total wins...at least that's my opinion.

Yertle wrote:

One last question, this only affects when 2+ players have the same number of wins right?  Ie a player with 7 wins and 2 losses wouldn't somehow ended up losing to a player with 6 wins and 3 losses (due to this SOS/SOSOS stuff), right?

Correct.  First check the #wins, then do the other tie breaker stuff.

One last last question, this would mean tournament scores wouldn't be needed at the individual tournament level at all right?

Correct.

I think it HAS to be W/L record, H-2-H, then whatever cockamamy tiebreaker method is developed.

I have been following this as well and really wanted to chime in, but haven't had the time to contribute.  I agree with conan and weather top (Wins, H2H, cockamamy) should be used to decide the tournament.  I think this is perfect for RR tournaments and the chances of a tie should be near zero.

Swiss tournaments are harder.

When I was thinking about this before, the one issue I had with Swiss tournaments it that all of the players need to be totally ranked each round.  To illustrate the problem, let look at 15 players in a 3 player per game Swiss Tournament.

Round 1: 5 winners, 10 losers - There is no way to distinguish the losers of the first round, yet one of them gets to play in a game where there are 2 winners.

Round 2:
  1 or 2 players have 2-0 record
  8 or 6 players have 1-1 record
  6 or 7 players have 0-2 record

Looking at the case of 2 players with a 2-0 record, 1 of the 6 players with a 1-1 record needs to be selected to play in the final round with the 2-0 guys.  I am curious as to how SODOS would deal with this?

It may be that there is just not enough information in the first two rounds to make the decision and then it must be made randomly and I except this, but was wondering if you have looked at this conan/hugh?

By the way Conan, thanks for doing the calculations on those tournaments, it is nice to see how the system would work if adopted.

Yertle wrote:

Is there a drawback to the Conan system?  Other than the slight possibility of still a pure tie?  (In which there probably isn't a good system is there?  Even manual wouldn't be able to resolve some ties.)

It depends on the values and needs of the community. I played in a lot of Swiss style chess tournaments in the 90s, and though the seeding was systematic, any tie in score at the end led to the prize money being split up evenly among the tied. No SOS/SOSOS. 

Summing the opponents' scores is nice because it gives an objective measure of how difficult the schedule was for each player. However, this is relative to the tournament being played. Imagine you beat BlackDog in a brilliant victory and then he proceeds to get terrible rolling for the rest of the tournament. He scores 0 points for the tournament and you get nothing towards your SOS (which is supposed to measure strength of schedule!).

For RR, summing only the defeated opponents score (since strength of schedule is, by default, equal) tips the scale towards rewarding beating good opponents more than punishing losing to bad ones. Some purists don't like this, but it breaks ties according to a criterion many find acceptable.

Also, sometimes the win record is perceived as "weighing much much more" than any of these statistical measures about who had the harder schedule or who beat the best opponents. In such cases, a group (such as the United States Chess Federation) will decide that ties are preferred over complaints of "Why did he win and not me?" (especially with money involved...)

Anyway, those are the types of criticisms that I am aware of. In spite of them, I perceive the world domination gaming community to be much more adverse to ties and much more willing to give it over to a statistic, as long as it does something like measure strength of schedule or of defeated opponents. 

So, I think Conan's system is a good one for this particular community.

Alpha, tiebreakers are typically not used to determine seeding. Your example gives a good reason why not.

The price you pay for the flexibility of a Swiss style tournament is that you have to match people from different brackets. Often those people are indistinguishable from the point of view of the tournament from others in their bracket. There is no way around this aside from using a different tournament style.

My guess is that tom uses T-score for seeding. It is typical to use outside information (USCF or FIDE ratings in a chess tournament for example) to determine the seeding of each round when the brackets don't align. 

(*) As a side note, it occurred to me that scheduled tournaments with their unpredictable registration numbers are the best application of the Swiss system. (the numbers argument makes sense too ... but if you've waited for 24 players, why not wait for 32 and have a bracket?)

Presumably that final round selection becomes less important / irrelevant with SODOS? Currently the winner of the top final round game is the tournament winner, with SODOS it will be calculated separately.

Conan wrote:

1. [Wins-->SODOS]  Ignore special H-2-H consideration.  Ties still possible (do whatever we're going to do with ties).
2. [Wins-->SODOS-->H-2-H]  Use H-2-H results as final tiebreaker.  You could still have a tie if both lost in their H-2-H (3 player game, for example).
3. [Wins-->H-2-H-->SODOS]  Apply H-2-H results before SODOS consideration.  Ties still possible.

I think any of the three options could be justified but most players probably wouldn't want to see the H-2-H winner lose.  I think options 2 and 3 are the most fair.  I think 3 sits best with me.

My out of the box offering:

4. [Wins-->SODOS-->SODODOS-->SODODODOS...] (Sum of the DO of the DO scores of the DO...).    Iterate until a winner emerges. 

I have no clue if this actually works, but it would give the system a consistent single-strand paradigm for determining a victor.

Hugh wrote:

My guess is that tom uses T-score for seeding.

I don't think he does, but have wondered if he should.

Hugh wrote:. 

(*) As a side note, it occurred to me that scheduled tournaments with their unpredictable registration numbers are the best application of the Swiss system. (the numbers argument makes sense too ... but if you've waited for 24 players, why not wait for 32 and have a bracket?)

So, are you suggesting that Swiss be eliminated.

I would suggest that the only problem with Swiss Tournament is that they will not always produce a clear winner.  It seems to me that after 3 (or whatever the default is for the number of players) rounds of a Swiss tournament, if no winner is clearly decided (i.e. many people have the same 2-1 record), then play one more round.  Use, number of wins, then comparisons, then SODOS to decide winner.

Any pure tie after this should be split (i.e. all tied players get the trophy) as it is just so unlikely to happen.

Although I will accept whatever is decided, I think the tournament deciding system should be site-wide and that tournament creators should not be given the choice.

Alpha wrote:

Hugh wrote:. 

(*) As a side note, it occurred to me that scheduled tournaments with their unpredictable registration numbers are the best application of the Swiss system. (the numbers argument makes sense too ... but if you've waited for 24 players, why not wait for 32 and have a bracket?)

So, are you suggesting that Swiss be eliminated?

No, almost the opposite. How many tournaments don't get off the ground because the tournament creator overestimated interest? More tournaments would start if you just said "join my 5 round Swiss style tournament, it will begin two weeks from now." To get the # of players be divisible by game size, you cut the last few people who joined. Have a minimum (set by tournament creator) in case you don't want a 5 round 8-player tournament on a 4-player map and have the max be capped by a full bracket (M^R or whatever). 

It seems like if you're guessing you can get 40 people and are willing to wait for the tournament to fill, you are more likely to say "let's make that 16 or 32 or 64 to guarantee a winner."

Alpha wrote:

I would suggest that the only problem with Swiss Tournament is that they will not always produce a clear winner.  It seems to me that after 3 (or whatever the default is for the number of players) rounds of a Swiss tournament, if no winner is clearly decided (i.e. many people have the same 2-1 record), then play one more round.  Use, number of wins, then comparisons, then SODOS to decide winner.

Playing more rounds than the minimum required to produce a winner at least seems more likely to tie than playing the minimum.  But I'm not so sure. You can get things like only two players remaining in the pure winners bracket on a 5 player map. If you are suggesting a playoff only among the tied, you can run into divisibility issues for multiplayer maps. Clever choices of size might sometimes solve the divisibility issue, but arbitrary choices of size doesn't work.

I like extending the tournament, or at least allowing a larger than the minimum number of rounds from the outset. The advantage of this is that even if a tie occurs, the SOS/SOSOS information becomes more defined.

Alpha wrote:

I would suggest that the only problem with Swiss Tournament is that they will not always produce a clear winner.  It seems to me that after 3 (or whatever the default is for the number of players) rounds of a Swiss tournament, if no winner is clearly decided (i.e. many people have the same 2-1 record), then play one more round.  Use, number of wins, then comparisons, then SODOS to decide winner.

Just to be clear:

• For RR it's       Wins-->H2H-->SODOS
• For Swiss its   Wins-->SOS-->SOSOS

Hugh wrote:

Playing more rounds than the minimum required to produce a winner at least seems more likely to tie than playing the minimum. 

Probably true. 

Hugh wrote:

I like extending the tournament, or at least allowing a larger than the minimum number of rounds from the outset. The advantage of this is that even if a tie occurs, the SOS/SOSOS information becomes more defined.

This is probably true.  But I wonder if you'd get cases of the top dogs having to face off more than once near the end.

Seeding.

Tom, can you explain how this is done currently?

My idea is to assign each player or team a seeding ranking.  This could be based on an average of their board ranking and their tournament ranking.  Or it could be 100% board rank or 100% tourney ranking.  I thought 100% board ranking might not be good because often many of the players haven't played the board yet, so it'd be more of a random seeding.  But it would be nice to see some influence of the board ranking, hence 50%-50%.  After the seeding rank is determined it does not change for the rest of the tourney.  After each round players would be lined up according to wins-->seeding and the Swiss pairings would proceed as normal.

Hugh wrote:

Alpha wrote:

Hugh wrote:. 

(*) As a side note, it occurred to me that scheduled tournaments with their unpredictable registration numbers are the best application of the Swiss system. (the numbers argument makes sense too ... but if you've waited for 24 players, why not wait for 32 and have a bracket?)

So, are you suggesting that Swiss be eliminated?

No, almost the opposite. How many tournaments don't get off the ground because the tournament creator overestimated interest? More tournaments would start if you just said "join my 5 round Swiss style tournament, it will begin two weeks from now." To get the # of players be divisible by game size, you cut the last few people who joined. Have a minimum (set by tournament creator) in case you don't want a 5 round 8-player tournament on a 4-player map and have the max be capped by a full bracket (M^R or whatever). 

It seems like if you're guessing you can get 40 people and are willing to wait for the tournament to fill, you are more likely to say "let's make that 16 or 32 or 64 to guarantee a winner."

I still think I am misunderstanding your post, but let's say I like the idea of a Swiss System tournament being a tournament that starts in x days with however many player have joined (trim last __ or wait for __ to join) so that there is a proper number of players per game.

Hugh wrote:

Playing more rounds than the minimum required to produce a winner at least seems more likely to tie than playing the minimum. 

Probably true. 

Hugh wrote:

I like extending the tournament, or at least allowing a larger than the minimum number of rounds from the outset. The advantage of this is that even if a tie occurs, the SOS/SOSOS information becomes more defined.

This is probably true.  But I wonder if you'd get cases of the top dogs having to face off more than once near the end.

I disagree that allowing more rounds would increase the likelyhood of a tie and as stated the SOS/SOSOS would be better with more rounds.  Also, the last round would have all of the players with the best record and score playing in the same game and thus the winner of that game would almost always win the tournament (often outright by record, rest of the time by SOS/SOSOS).