Suggestion on Swiss-System Scoring (BWR Spoilers)

VGhost

United States3606 Posts

August 25 2014 15:28 GMT

I greatly enjoyed the Brood War Reunion showcase tournament played over this past weekend. I want to thank and congratulate Artanis[Xp], Sayle and his co-casters, and all the players who put in their time and effort to make the event possible.

However, I also want to take this opportunity to discuss one of the drawbacks of the Swiss system for Starcraft tournaments. This is not a criticism of this tournament's organizers, as they followed the rules to the letter as far as I can determine.

Scoring

In the normal Swiss system, byes are credited as wins. Since the system was originally designed for chess, and the most comprehensive rules are still maintained by FIDE, this actually makes a certain amount of sense. Draws are very common in chess at all competitive levels; probably more common the higher the level of play. Since in general players are paired from highest to lowest, any byes will usually only affect the players with worse scores. This means that players getting byes - especially later byes - are unlikely to figure into any prizes given immediately or after a playoff; however, the extra points they are awarded put a little extra pressure on the top - and especially the middle - players to play for the win, and not settle for draws. Whether this is the official justification I do not know. It is an obvious and reasonable one to the casual observer.

For Starcraft, the case is a bit different. Draws are extremely unusual and are by convention replayed. Further, the final rounds of almost any tournament are, again by convention, a playoff, and it is unusual that any given stage of a tournament after declared preliminaries (if any) but before the final playoff will eliminate many more than half the competitors. In the case of this tournament, 8 of 19 advanced to the playoff; a harsher cut than many other events but still approximately half the field.

(By score, in fact, 14 players qualified for the eight spots (there being a 7-way tie for 8th). This is partly due to playing an extra round more than necessary to separate the competitors: 19 is greater than 16 but less than 32, so five rounds were called for, rather than six. After five rounds, ten players were in competition for eight spots, with seven holding a 3-2 score and the five 4th-8th spots to fill. This is not ideal but better than the situation after the sixth round, which allowed scores to even up again.)

I believe that for Starcraft, byes should simply be credited as no score, rather than crediting players with wins they did not achieve. Possible methods for resolving the imbalanced game results - players with byes will have one fewer games scored - include playing a final round between any players with byes; creating a smaller play-in pool for any players who might with a sixth game have made the playoff or have changed seeds; and not doing anything but taking the results as they stand.

I have included below the tournament table, redrawn by round according to this suggestion but keeping the actual game results. Since initial player seeds were not included on Liquipedia, I simply alphabetized the table rather than trying to work them out and do it exactly. In some cases, my proposed change would have also changed the pairing for a round; where this is true I have noted it. If I have set it up correctly, this would have only affected seven games (out of 54 played) and none of the byes (out of six given) and possibly not changed the final game scores at all.

Round One
+ Show Spoiler [Games] +

Arcneon 1-0 Lamer
MaNa 1-0 Arew
Mondragon 1-0 CastrO
ClouD 1-0 Nyoken
Eriador 1-0 PredY
Ptak 1-0 Fosken
SarenS 1-0 G5
Sen 1-0 GoOdy
HayprO 1-0 TT1
Socke no game

+ Show Spoiler [Standings] +

Arcneon 1-0
ClouD 1-0
Eriador 1-0
HayprO 1-0
MaNa 1-0
Mondragon 1-0
Ptak 1-0
SarenS 1-0
Sen 1-0
Socke 0-0
Arew 0-1
CastrO 0-1
Fosken 0-1
G5 0-1
GoOdy 0-1
Lamer 0-1
Nyoken 0-1
PredY 0-1
TT1 0-1

Round Two
+ Show Spoiler [Games] +

Mondragon 1-0 Arcneon
ClouD 1-0 Ptak
Eriador 1-0 SarenS
HayprO 1-0 Sen
MaNa 1-0 Socke
Arew 1-0 GoOdy
CastrO 1-0 Lamer
Fosken 1-0 Nyoken
G5 1-0 PredY
TT1 no game

+ Show Spoiler [Standings] +

ClouD 2-0
Eriador 2-0
HayprO 2-0
MaNa 2-0
Mondragon 2-0
Arcneon 1-1
Arew 1-1
CastrO 1-1
Fosken 1-1
G5 1-1
Ptak 1-1
SarenS 1-1
Sen 1-1
Socke 0-1
TT1 0-1
GoOdy 0-2
Lamer 0-2
Nyoken 0-2
PredY 0-2

Round Three
+ Show Spoiler [Games] +

MaNa 1-0 HayprO
Mondragon 1-0 ClouD
Ptak 1-0 Socke*
Sen 1-0 Arcneon
G5 1-0 Arew
GoOdy 1-0 Nyoken
SarenS 1-0 Fosken
TT1 1-0 Eriador*
PredY 1-0 CastrO*
Lamer no game

* Games would have been (e.g.): Eriador vs. Ptak, CastrO vs. TT1, Socke vs. PredY. Assume Ptak 1-0 Eriador, TT1 1-0 CastrO, PredY 1-0 Socke, and the resulting score is the same.

+ Show Spoiler [Standings] +

MaNa 3-0
Mondragon 3-0
ClouD 2-1
Eriador 2-1
G5 2-1
HayprO 2-1
Ptak 2-1
SarenS 2-1
Sen 2-1
TT1 1-1
Arcneon 1-2
Arew 1-2
CastrO 1-2
Fosken 1-2
GoOdy 1-2
PredY 1-2
Lamer 0-2
Socke 0-2
Nyoken 0-3

Round Four
+ Show Spoiler [Games] +

MaNa 1-0 Mondragon
ClouD 1-0 Sarens
Eriador 1-0 Sen
TT1 1-0 G5
HayprO 1-0 Ptak
Arcneon 1-0 Socke*
Lamer 1-0 Arew*
Fosken 1-0 CastrO
PredY 1-0 GoOdy
Nyoken no game

* Games would have been Arcneon vs. Arew, Lamer vs. Socke. Assume Arcneon 1-0 Arew, Lamer 1-0 Socke, and the resulting score is the same.

+ Show Spoiler [Standings] +

MaNa 4-0
ClouD 3-1
Eriador 3-1
HayprO 3-1
Mondragon 3-1
TT1 2-1
Arcneon 2-2
Fosken 2-2
G5 2-2
Ptak 2-2
PredY 2-2
SarenS 2-2
Sen 2-2
Lamer 1-2
Arew 1-3
CastrO 1-3
GoOdy 1-3
Nyoken 0-3
Socke 0-3

Round Five
+ Show Spoiler [Games] +

MaNa 1-0 Eriador
TT1 1-0 Mondragon
SarenS 1-0 Sen
PredY 1-0 Fosken
HayprO 1-0 ClouD
Ptak 1-0 Lamer
G5 1-0 Arcneon
CastrO 1-0 Socke*
Nyoken 1-0 Arew*
GoOdy no game

* Games would have been Arew vs. CastrO, Nyoken vs. Socke. Assume Castro 1-0 Arew, Nyoken 1-0 Socke, and the resulting score is the same.

+ Show Spoiler [Standings] +

MaNa 5-0
HayprO 4-1
TT1 3-1
ClouD 3-2
Eriador 3-2
G5 3-2
Mondragon 3-2
PredY 3-2
Ptak 3-2
SarenS 3-2
Arcneon 2-3
CastrO 2-3
Fosken 2-3
Sen 2-3
GoOdy 1-3
Lamer 1-3
Nyoken 1-3
Arew 1-4
Socke 0-4

Round Six
+ Show Spoiler [Games] +

TT1 1-0 MaNa
Mondragon 1-0 HayprO
ClouD 1-0 PredY
Eriador 1-0 G5
SarenS 1-0 Ptak
Fosken 1-0 Arcneon
Castro 1-0 Sen
GoOdy 1-0 Lamer
Nyoken 1-0 Socke
Arew no game

+ Show Spoiler [Standings] +

MaNa 5-1
TT1 4-1
ClouD 4-2
Eriador 4-2
HayprO 4-2
Mondragon 4-2
SarenS 4-2
Fosken 3-3
CastrO 3-3
G5 3-3
PredY 3-3
Ptak 3-3
GoOdy 2-3
Nyoken 2-3
Sen 2-4
Arcneon 2-4
Lamer 1-4
Arew 1-4
Socke 0-5

Due to byes, TT1 (4-1), GoOdy (2-3), Nyoken (2-3), Lamer (1-4), Arew (1-4), and Socke (0-5) had only played five games. It would be possible to let these be the final standings. One benefit is that it leaves only five players in competition for the 8th playoff spot; the drawback is of course that two players have some kind of a claim of being slighted by not playing a sixth game.

If a final round to make up that sixth game for these players is played, the games would be (e.g.): TT1 vs. GoOdy, Nyoken vs. Lamer, Arew vs. Socke. Note that there is no way for the actual scores to match the scores assigned by calling byes "wins". This is obvious anyway, as that resulted in 6 wins assigned to 0 losses, imbalancing the table. Particularly, by actually playing the sixth games, it is not possible that both TT1 ties for 1st and GoOdy and Nyoken tie for 8th. If TT1 1-0 GoOdy, GoOdy is eliminated; if GoOdy 1-0 TT1, TT1 drops to 4-2 and a tie for 2nd. Meanwhile Nyoken can play himself in or out of the 8th place tie.

Alternatively - though less ideal - we can shorten the finishing round to one game. Lamer (1-4), Arew (1-4), and Socke (0-5) are mathematically eliminated. This leaves three players to be determined. TT1 is guaranteed a spot in the playoff, but a sixth game could change his seeding (tied for 1st vs. tied for 2nd). GoOdy and Nyoken (2-3) could each by playing a sixth game join the tie for 8th if they won. Since GoOdy defeated Nyoken head-to-head in Round Three, Nyoken could be eliminated from consideration, and TT1 vs. GoOdy played to determine final seeding. The complaint about this is obvious. If GoOdy loses - likely, in this case - his final record stands at 2-4, while Nyoken would have a 2-3 record but would not have been given a chance to play in to the 8th place tie.

At any rate, several players will still have the same record, and the participants in the playoff, and seeding, ought to be determined - I would say preferably by non-random methods.

Tie-breaking

Formally, ties are normally broken by any previous head-to-head play and some variant of the Buchholz (or, Solkoff) system - in which players with equal scores are ranked by some sum of their opponents's score. ELO or other ratings are sometimes used when applicable, although that is not pertinent here.

The final table (as recorded on Liquipedia) was:

1. TT1 5-1
2. MaNa 5-1
3. Mondragon 4-2
3. HayprO 4-2
3. Eriador 4-2
3. SarenS 4-2
3. ClouD 4-2
8. Fosken 3-3
8. G5 3-3
8. PredY 3-3
8. Ptak 3-3
8. CastrO 3-3
8. GoOdy 3-3
8. Nyoken 3-3
15. Sen 2-4
15. Arcneon 2-4
15. Lamer 2-4
15. Arew 2-4
19. Socke 1-5

A FFA was suggested as a tie-breaker for the 8th spot, in keeping with the light-hearted spirit of the tournament, but eventually Fosken was randomly drawn. Using instead the simplest possible Buchholz system counting opponents' wins, the table would be tie-broken as follows:

1. TT1 5-1, 1-0 MaNa
2. MaNa 5-1, 0-1 TT1
3. Mondragon 4-2, 23 (Castro 3, Arcneon 2, ClouD 4, MaNa 5, TT1 5, HayprO 4), 1-0 HayprO
4. HayprO 4-2, 23 (TT1 5, Sen 2, MaNa 5, Ptak 3, ClouD 4, Mondragon 4), 0-1 Mondragon
5. Eriador 4-2, 22 (PredY 3, SarenS 4, TT1 5, Sen 2, MaNa 5, G5 3)
6. ClouD 4-2, 21 (Nyoken 3, Ptak 3, Mondragon 4, SarenS 4, HayprO 4, PredY 3)
7. SarenS 4-2, 19 (G5 3, Eriador 4, Fosken 3, ClouD 4, Sen 2, Ptak 3)
8. G5 3-3, 20 (SarenS 4, PredY 3, Arew 2, TT1 5, Arcneon 2, Eriador 4), 1-0 PredY
9. PredY 3-3, 20 (Eriador 4, G5 3, CastrO 3, GoOdy 3, Fosken 3, ClouD 4), 0-1 G5
10. Ptak 3-3, 18 (Fosken 3, ClouD 4, Socke 1, HayprO 4, Lamer 2, SarenS 4), 1-0 Fosken
11. Fosken 3-3, 18 (Ptak 3, Nyoken 3, SarenS 4, CastrO 3, PredY 3, Arcneon 2) 0-1 Ptak
12. Castro 3-3, 15 (Mondragon 4, Lamer 2, PredY 3, Fosken 3, Socke 1, Sen 2)
13. Nyoken 3-3, 13 (ClouD 4, Fosken 3, GoOdy 3, BYE 0, Arew 2, Socke 1)
14. GoOdy 3-3, 12 (Sen 2, Arew 2, Nyoken 3, PredY 3, BYE 0, Lamer 2)
15. Sen 2-4, 20 (GoOdy 3, HayprO 4, Arcneon 2, Eriador 4, SarenS 4, CastrO 3)
16. Arew 2-4, 16 (MaNa 5, GoOdy 3, G5 3, Lamer 2, Nyoken 3, BYE 0)
17. Arcneon 2-4, 15 (Lamer 2, Mondragon 4, Sen 2, Socke 1, G5 3, Fosken 3)
18. Lamer 2-4, 13 (Arcneon 2, CastrO 3, BYE 0, Arew 2, Ptak 3, GoOdy 3)
19. Socke 1-5

This could obviously be applied to my hypothetical scoring table above, as well. A slightly more complicated variation on the system (Harkness or Median-Buchholz) drops the highest and lowest opponents' scores, which helps correct for any early-round mismatches as well as byes. In this case, byes are not usually counted as wins for tie-breaking. For example, Mondragon's tie-breaker would be 15: TT1 scores as 4 since he had a bye and then Arceneon's 2 and MaNa's 5 are dropped.

Based on the actual tie-breaker calculated above, the resulting brackets would have been:

1. TT1 vs. 8. G5
4. HayprO vs. 5. Eriador

3. Mondragon vs. 6. ClouD
2. MaNa vs. 7. SarenS

These are identical to the actual games except for the inclusion of Fosken over G5. However, in the actual tournament, HayprO and Eriador were in the lower bracket, while Mondragon and ClouD were in the upper one. I am assuming a random draw was made, unless Artanis[Xp] has a further explanation.

Obviously for the Reunion tournament all of this is relatively trivial, as the main point was to have fun and put on a show, and both of those goals were accomplished brilliantly. I thought some people might be interested in some of the further details behind the system, however. I had run a tournament for the chess club in college, but to refresh my memory I did go check on FIDE's website and Wikipedia - I do not by any means remember all this stuff myself without writing it down somewhere. Any errors in calculation or mis-representation of the systems involved are entirely my own fault.

Artanis[Xp]

Netherlands12968 Posts

August 26 2014 01:53 GMT

I used the software "Swiss Perfect" to determine the matchups for the most part. Unfortunately I couldn't enter results for past rounds if I started a new round, so to start new rounds I had to manually assign matchups, remember them and hope it worked out. It was not ideal.

As for the tiebreakers, these were the stats Swiss Perfect gave me on it. The first score after the games is M-Buch., the second is Buch., and the third is Progr. I'm not sure what these all mean, but perhaps you could shed some light on it.
[image loading]

The playoffs were decided based on ranking in the way you presented, but I must've made a mistake and accidentally put the HayprO vs Eriador match in the lower bracket instead of the upper one as there was a lot of pressure for time with how the FFA turned out.

The reasons why I picked the system was because it was one of the only systems I could think of that could account for as many players as could sign up, and because I expected a large skill differential so I wanted to allow the best players to come through. I think it succeeded in that regard pretty well despite the random draw. I like the idea regarding not counting BYEs and then having an extra round though. I'll keep that in mind.

Foolishness

United States3044 Posts

August 26 2014 06:54 GMT

On August 26 2014 00:28 VGhost wrote:
In the normal Swiss system, byes are credited as wins. Since the system was originally designed for chess, and the most comprehensive rules are still maintained by FIDE, this actually makes a certain amount of sense. Draws are very common in chess at all competitive levels; probably more common the higher the level of play. Since in general players are paired from highest to lowest, any byes will usually only affect the players with worse scores. This means that players getting byes - especially later byes - are unlikely to figure into any prizes given immediately or after a playoff; however, the extra points they are awarded put a little extra pressure on the top - and especially the middle - players to play for the win, and not settle for draws. Whether this is the official justification I do not know. It is an obvious and reasonable one to the casual observer.

Fyi, in chess "full-point byes" can only occur in standard swiss tournaments (FIDE rule) and only occur if there are an odd number of players in a round. This is different from half-point byes which you can usually request and zero-point byes (though for the BW tournament we are looking at this would be a case for a full-point bye). When assigning a full-point bye in chess it goes to the lowest-rated player with the lowest score who has not already received a full-point bye.

I don't know if that's the official justification for the full-point bye. I always thought it was because it sucks to not have to play a round.

VGhost

United States3606 Posts

August 26 2014 13:45 GMT

On August 26 2014 15:54 Foolishness wrote:
Fyi, in chess "full-point byes" can only occur in standard swiss tournaments (FIDE rule) and only occur if there are an odd number of players in a round. This is different from half-point byes which you can usually request and zero-point byes (though for the BW tournament we are looking at this would be a case for a full-point bye). When assigning a full-point bye in chess it goes to the lowest-rated player with the lowest score who has not already received a full-point bye.

I don't know if that's the official justification for the full-point bye. I always thought it was because it sucks to not have to play a round.

Yes - obviously this is all a moot point if there are an even number of players in the tournament. I don't fully follow your "half-point byes" and "zero-point byes" and can't find a reference to them anywhere on FIDE's website - could you elaborate? A "zero-point bye", I would guess, is a forfeited game; I can't figure out a "half-point bye" at all.

On August 26 2014 10:53 Artanis[Xp] wrote:
I used the software "Swiss Perfect" to determine the matchups for the most part. Unfortunately I couldn't enter results for past rounds if I started a new round, so to start new rounds I had to manually assign matchups, remember them and hope it worked out. It was not ideal.

As for the tiebreakers, these were the stats Swiss Perfect gave me on it. The first score after the games is M-Buch., the second is Buch., and the third is Progr. I'm not sure what these all mean, but perhaps you could shed some light on it.
[image loading]

The last column, "Progr.", seems to be a progressive rating. The method is set up to reward wins earlier in the tournament, based on the theory that it's easier to win and so "come back" in the later rounds. Although the program documentation says this is "common knowledge", it doesn't explain why, though it's pretty obvious: a player at 2-3 will likely get an easier opponent than one at 3-2. Personally, since opponents' records are accounted for in other systems; I find intentionally "correcting" for this problem to be perhaps not quite legitimate. The score is determined from a table based on the order of a player's results; however, I can't seem to find a table anywhere. It seems to be included in the printed version of the official chess rules, but I don't have that.

The "Buch." and "M-Buch." columns are the Buchholz and Median-Buchholz scores I referenced above.

I'm unsure why the Buchholz numbers are different than mine. The scores that are lower seem to have counted opponents' byes as only 0.5 points, as suggested on Wikipedia for the Median-Buchholz. I don't see any suggestion that that's usually done for the simple Buchholz; I know I didn't do that yesterday writing this. It would change some of the scores. The scores that are higher than my calculations (Nyoken, GoOdy) I can't figure out at all.

Median-Buchholz is then calculated the same way (byes as 0.5 points for opponents), but with the highest and lowest not counted. The numbers seem consistent with the Buchholz numbers generated by the program, but again Nyoken and GoOdy's numbers don't match my calculations.

An aside: there's something a little bit fishy about a tie-breaking system that's going by players' scores but using a different scoring system than the actual tournament. I understand, of course, why it's not giving full credit for a bye "win" - the main point of my post, after all, was that this distorts standings - but to award those points in the main table scoring, and then not award them when doing tie-breakers, seems a little odd.

Foolishness

United States3044 Posts

August 27 2014 19:15 GMT

On August 26 2014 22:45 VGhost wrote:

Show nested quote +

A player can request a half-point bye for a round of his choosing, meaning the player does not play a game that round and is awarded half a point instead. Though I'm not sure about the origin, I believe the primary reason it's there is to make it easier on the players (e.g. I can't play the last round of this tournament because I have to catch my flight home). A zero-point bye is not playing a game and not getting any points for it (which is different from forfeiting your game because a zero-point bye means you're not matched up against another player).

The rules for requesting a half-point bye are usually pretty strict to avoid situations where there's incentive for a player to take one (for the purpose of winning prizes). For example, in a lot of big tournaments I've played in you have to declare your bye at least two rounds before the round you intend to take it. I've seen some tournaments where they won't even allow you to take a half-point bye in the last half of the tournament. Depending on how many rounds the tournament is there's a maximum cap on how many you're allowed to take (usually 2). The USCF's Official Rules of Chess has all of this documented and more.

I'm not sure what FIDE says about half-point byes (if it's even allowed). I can't find anything on it either x.x

Please or register to reply.

Suggestion on Swiss-System Scoring (BWR Spoilers)

Completed

Ongoing

Upcoming