Controversial pairings from SwissSys

Running Chess Tournaments
1

In a recent tournament, SwissSys generated round 4 pairings that caused quite a bit of controversy.

Here is an ASCII formatted version of the wallchart after three rounds. If this isn’t particularly readable, there’s an HTML version at http://www.mv.com/ipusers/crab/wallchart.html

Name Rtng Rd 1 Rd 2 Rd 3 Tot 1 Alpha 2638 W 10 B 6 W 3 1.0 2.0 3.0 3.0 2 Bravo 2535 B 11 W 5 H--- 1.0 1.5 2.0 2.0 3 Charlie 2508 W 12 B 9 B 1 1.0 1.5 1.5 1.5 4 Delta 2410 B 13 W 18 B 7 0.5 1.5 2.0 2.0 5 Echo 2407 W 14 B 2 W 9 1.0 1.5 2.5 2.5 6 Foxtrot 2302 B 15 W 1 W 17 1.0 1.0 2.0 2.0 7 Golf 2287 H--- B 16 W 4 0.5 1.5 2.0 2.0 8 Hotel 2244 W 16 H--- B 13 0.5 1.0 1.5 1.5 9 India 2236 B 17 W 3 B 5 1.0 1.5 1.5 1.5 10 Juliet 2232 B 1 W 13 B 16 0.0 0.5 1.5 1.5 11 Kilo 2200 W 2 B 14 W 15 0.0 0.0 0.0 0.0 12 Lima 2119 B 3 W 15 B 19 0.0 1.0 2.0 2.0 13 Mike 2086 W 4 B 10 W 8 0.5 1.0 1.5 1.5 14 November 2069 B 5 W 11 H--- 0.0 1.0 1.5 1.5 15 Oscar 2048 W 6 B 12 B 11 0.0 0.0 1.0 1.0 16 Papa 2037 B 8 W 7 W 10 0.5 0.5 0.5 0.5 17 Quebec 2015 W 9 bye B 6 0.0 1.0 1.0 1.0 18 Romeo 1967 bye B 4 H--- 1.0 1.0 1.5 1.5 19 Sierra 1832 H--- H--- W 12 0.5 1.0 1.0 1.0

(I’m not sure why the label “Code” appears before the wallchart, but no matter.)

Note that player 8 (Hotel) is taking a half point bye in round 4.

In particular, the controversy was caused by how SwissSys handled the 2.0 score group. Here are the pairings SwissSys version 6.084 generated:

Echo - Alpha Delta - Bravo Lima - Foxtrot Charlie - Golf India - Mike Juliet - November Romeo - Quebec Oscar - Sierra Kilo - Papa

There are five players in the 2.0 score group: Bravo, Delta, Foxtrot, Golf, and Lima.

I believe the “natural” pairings would have called for dropping Lima down to the 1.5 score group. Since Delta (4) and Golf (7) have already played, we would transpose Foxtrot and Golf to produce:

Bravo - Golf Delta - Foxtrot

Since Lima (12) has already played Charlie (3) and Hotel ( 8 ) has a half point bye, we then pair Lima and India. (Both are due white, and both have identical color history, so Lima gets white.)

So, the “natural” pairings seem to work out to be:

Bravo - Golf Delta - Foxtrot Lima - India Charlie - Mike Juliet - November

In terms of color allocation, there are two “problems” with the above. First, the pairing Bravo - Golf violates color alternation for Bravo. Second, the pairing Lima - India violates color equalization for India.

After a long discussion (more than a half hour) among the three TDs (an NTD and two senior TDs), we realized that the SwissSys pairings had the virtue of making the colors work correctly in all five pairings. It appears that SwissSys chose to make a 168 point transposition to treat Golf instead of Lima as the odd man in the 2.0 score group in order to improve the color allocation.

The discussion then turned to whether SwissSys should have done so. Rule 29D1a does include the text “Care must be taken … that the color consequences are acceptable (29E, 29E3).”

As they used to say in college: “Discuss.” Was SwissSys correct here? How would experienced TDs handle this?

2

Two questions Ken:

  1. Was this the final round?

  2. Did Swis-Sys know how many rounds there were in the event?

3

Would this matter? Some TDs treat last-round color allocations differently, but this is a variation.

Without actually setting this up with pairing cards, it appears from the description that the 168-point transposition allowed the equalization of color on at least one more board. This is well within the 200-point limit of 29E5b. The question is to what extent this applies to dropping the odd man, a point which has always been unclear in the rules.

4

It might have made a difference to Swis-Sys, John.

5

I suppose it might, but that would mean that a) Thad Suits incorporated a subroutine which, as far as I know, is not included in the rules, and b) the program would produce different pairings depending on whether you specified the number of rounds in the Setup menu (which is optional, not required). If the latter is the case, it’s a bug that needs squashing, but I have no reason to think that it is. (Incidentally, setting the number of rounds does not prevent the program from pairing additional rounds, so I suspect this field is purely cosmetic.)

6

There was a minor controversial issue with SwissSys 6.0. It was not the issue of color history or the pairings, it was the assignment of the board numbers. During the third round, a Class B player had a draw against the Master. When the pairing went up for the fourth round, the Class B player was placed onto board one, the Master was placed onto board two. The Class B and the Master had equal point numbers (3.5) going into the last round. They were the only players with 3.5 going into the last round, the Master was questioning why he was bumped down from board one to board two.

7

I have had several discussions on the topic of how the program pairs with the author. As he points out the TD, not the program, is in charge. The TD can simply switch the pairings. One of the things most TDs at the large events, that I have had the joy to work at, do is to check those pairings over before they get posted. If they don’t agree with them, they change them.

The real question here is a matter of “best” pairings. Both sets appear legal.

Tim

8

  1. No, this was the fourth round of six.

  2. Yes, or at least I think it did. When I set up the section, I specified six rounds for the section.

9

Be careful when you say “natural” – that’s an emotionally charged word. What looks “natural” when you focus on one tiny point in the pairing list may look highly unnatural when you focus elsewhere.

You may be overlooking that, when the lowest-rated 2.0 has already played the highest-rated 1.5, there are TWO alternatives:

Alternative A: Pair the lowest 2.0 against a 1.5 other than the highest (such as the second-highest).

Alternative B: Pair a 2.0 other than the lowest (such as the second-lowest) against the highest 1.5.

The two alternatives are equally plausible, and equally “natural”. Which is better depends on the circumstances.

In this case, alternative A (pairing Lima against India rather than against Charlie) is a 272-point transposition. Alternative B (pairing Charlie against Golf rather than against Lima) is a 168-point transposition.

It is likely that Swis-Sys chose alternative B not “in order to improve the color allocation” but rather because it was the better of the two ways (smaller transposition) to avoid pairing the same players twice. The better colors were just a bonus.

So, the dropping of Golf rather than Lima should not have been controversial. If there was to be controversy, it should have been over the way Swis-Sys handled the remaining four 2-pointers.

The “raw” pairings would have been:

Bravo - Foxtrot Delta - Lima
– but this would result in bad color alternation on one board and bad color equalization on the other. There are two possible transpositions:

Transposition X:

Bravo - Lima Delta - Foxtrot
– and transposition Y:

Bravo - Delta Foxtrot - Lima
– both of which solve these problems. Transposition X is 125 points, while transposition Y is 108 points. Swis-Sys seems to prefer the latter, even though it involves an interchange – switching a player just above the middle of the group with one just below.

The rulebook has the following (all on page 150) to say about transpositions vs interchanges:

  1. “While interchanges are sometimes necessary, they should not be used if adequate transpositions are possible.” This seems to support transposition X over transposition Y, at least if “adequate” means “within 80 points to alternate or 200 points to equalize”.

  2. "A transposition that satisfies … the 80-point rule, should be preferred to any interchange … " This has no bearing on the case above, since transposition X is over 80 points.

  3. “If pairing a round in which … the 200-point rule, is used … an interchange involving a smaller rating switch than a transposition should be preferred to the latter unless the transposition satisfies the 80-point rule.” This seems to support transposition Y, and to contradict 1. above.

Swis-Sys seems to have gone along with transposition Y, justified by 3. above. Many human TDs, however (myself included), might feel more comfortable with transposition X.

Don’t tell me, let me guess. The loudest squawk came from Delta, right? Some players seem to assume that, just because they are above the middle of their score group by a few points, they have a God-given right to a lower-rated opponent in the next round. Not so. If you’re in the top half of your score group, but within 80 or 200 points of somebody in the bottom half, or vice versa, you’re transposable with that other person, and you could be paired against either the highest or the lowest player in the group.

Bill Smythe

10

Having said all of the above, if I were the TD (or the pairing program), I think I might have dropped Golf rather than Lima even if Lima HADN’T already played Charlie. It makes the colors better in two score groups.

To look at it another way, the players in the 2.0 group (in rating order) are:

due-black due-white due-black due-black due-white
– while in the 1.5 group we have:

due-white due-white due-white due-black due-black due-white
Since there is an extra due-black in the 2.0 group, and extra due-whites in the 1.5 group, it makes sense to pair a 2.0 due-black against a 1.5 due-white. It improves the colors in both score groups. This is exactly what Swis-Sys did.

Bill Smythe

11

Doug:

How on earth did the players have 3.5 going into the last round of a 4-round tournament? Either you meant 2.5, or the tournament was a 5-rounder.

I also have six questions:

  1. What was the B player’s exact rating?
  2. What was the master’s exact rating?
  3. What was the rating of the B player’s last-round opponent?
  4. What was the rating of the master’s last-round opponent?
  5. What was the score (going into the last round) of the B player’s last-round opponent?
  6. What was the score (going into the last round) of the master’s last-round opponent?

Bill Smythe

12

To exactly the same extent it applies to pairings within a score group.

Under 29D, “The odd player”, it says “… switches to correct colors should stay within the appropriate limits (29E5).” 29E5 consists of 29E5a (the 80-point rule) and 29E5b (the 200-point rule).

Bill Smythe

13

On the other hand, the current versions of Swis-Sys and WinTD are getting so sophisticated at making pairings, that almost never can a human improve on them. Usually, if a human thinks he sees a better pairing, he is overlooking something, or looking only at a narrow question, such as a single pairing, rather than at the big picture, in a way only a program can do.

It’s gotten to the point where even the best NTDs ought to think really, really hard before they overrule pairings produced by the programs.

Bill Smythe

14

The way I read the rules the 80/200 point limits don’t enter into deciding which odd man to drop and play which player. I think the way I would program it to follow 29D is to compare the 2 lowest players in the 2.0 scoregroup (Golf/2287 and Lima/2119 - a 168 point difference) and the 2 highest players in the 1.5 scoregroup (Charlie/2508 and India/2236 - a 272 point difference) and to pick the one with the smallest impact.

And that’s the way I would do it without the rulebook - Charlie is SO much higher rated than the others in his score group that he “should” be the one that plays up.

15

Yeh, I think that’s a good way of looking at what I said.

Yup, that’s where the 272 comes from.

Here I do not agree. In deciding whether it’s legal to pair down player Y (the second-lowest) instead of player Z (the lowest) just to improve colors, one can look at the rating difference between players Y and Z. If it’s 200 or less, you can make this transposition to improve color equalization. If it’s 80 or less, you can make it to improve color alternation.

Look at 29D (“The odd player”), part of which says “switches to correct colors should stay within the appropriate limits (29E5)”, where 29E5 is the 80-point rule and the 200-point rule.

Bill Smythe

16

Also, if the SwissSys settings are set to FIDE defaults, it will do whatever it can to improve colors, regardless of ratings. We had a situation a few months ago when a master was paired down into the next scoregroup to improve colors and nobody could figure out why.

Alex Relyea

17

Actually, no. By far the loudest complaint came from Bravo.

18

I’ve usually found that when WinTD makes strange pairings, it is related to the settings made governing the pairings.

The last time I was ready to override a pairing it made in an adult tournament, I did the manual pairing and then the person at the computer looked at what I did and changed a setting that allowed WinTD to make the same pairing.

The setting he changed was the one that allowed interchanges within 80 points between the top and bottom halves. He had it set to prohibit interchanges entirely.

Since then, whenever I see a strange WinTD pairing I’ll first check the settings, change them, and redo the pairings. I’m guessing SwissSys users would say the same about SwissSys.

As far as your concern about the other pairings go, the setting may have been 200/200 for interchanges/transpositions, rather than 80/200.

One concern for the future is that there is an increasing percentage of TDs with only minimal experience doing manual pairings, so problems with program settings are getting less likely to be caught.

19

As for the first part (where I repeated what you said) - sorry. Too many posts to digest at once is my excuse …

But the second one is making me think some more. I tend to view the odd player stuff as more ‘absolute’ than color issues if that makes sense. In other words, I’ll drop the player the rule book says to, and then worry about colors later. The only reason (I think) that I would go back and re-think that is if someone is then forced to get 3 blacks/whites in a row which I also regard as an absolute… (but less of one than the odd player?)

You seem to advocating a more holistic view of your satisfaction with the pairings than I am able to at the moment. To me you’re just opening yourself up for more people to complain about the pairings. I guess I do as little as possible … just because a more-complicated (which means something I can’t define yet) pairing is legal doesn’t mean it’s OK …

It does make me hope that your point-count pairing rules will be clearly defined by the time the next edition comes out :slight_smile:

Thanks

20

Boo. The highest-rated player in the score group belly-ached because he had to play the second-highest instead of the third-highest? Cluck, cluck.

Bill Smythe