What is the correct pairing?

Running Chess Tournaments
1

Please help me determine the correct pairing for this situation.

Fifth and final round of minor GP tournament, class pairings certainly not an issue.

Players, Score, Rating, Rds 1-4 color history.
A, 4.0, 2350, unsure
B, 3.5, 2550, unsure
C, 3.5, 2130, BWBW
D, 3.5, 1940, WBWB
E, 3.5, 1930, BWBW
F, 3.5, 1900, WBWB
G, 3.5, 1890, WBWB

Note that Player C has played Player F. Also none of C-G have played anyone in the next lowest score group.

SwissSys paired A & B, which is not in dispute. Then followed D (w) vs. C (b), F(w) vs. E(b) with G-dropping.

Now, as I see it, within the group C-F the obvious transposition of E & F for color alteration doesn’t work because C & F have already played. My question is why didn’t the program allow the transposition of F & G (within 80 points), when F drops down instead, and then simple pairings follow? I even think ignoring color alteration (with C-E, D-F) is better since colors are already equal throughout and the basic 27A3 pairing principal would not be violated. Why did the computer decide to overvalue general pairing principal 27A5 (color alternation) over 27A3 Upper half vs. Lower Half?

My gut TD instincts say that this rule is further exacerbated for the odd-numbered final round when who could care less who gets what color one extra time over someone else - just don’t have the top half playing the rest of the top half etc.

I’ve shown this locally to several TDs including 2 nationals and everyone has agreed that the computer’s pairing seems strange. In the actual game, I was player D, and filed a protest the instant the ink was visible from the printer. The local TD/organizer was disinclined (I would say adamently opposed) to change the pairing. When a special referee was called, he eventually decided it was a legal, but not great pairing: “Not what I would necessarily do” was his his best euphamism. Fortunately I won anyway (with half my time to start off with), but that misses the point with respect to the Truth of the matter.

Is SwissSys wrong? Or do I need a pairing lesson?
Thanks for your input, Ben Bentrup

2

I don’t agree with the pairing, but I see why the program made it. Take a look at 29D1 a) and b). What these seem to say is that you should first determine the odd man. This should be the lowest-rated player in the scoregroup unless you cannot make acceptable pairings among the players that remain. If that happens, you try another iteration and see if dropping someone else helps. In this case, there was a perfectly legal pairing after dropping player G. A human would notice that there was something odd about the result, but a computer program has to follow a specific set of procedures.
As a TD, I would probably have overruled the computer and changed the pairing, but it is not “wrong” under the rules as written.

3

The Swis-Sys pairing is less strange than it first appears. It is an example of an “interchange” as opposed to a “transposition”. (See the 5th edition rulebook, rule 29E5 and the discussion following.) The idea is to switch a player at (or near) the bottom of the top half, with a player at (or near) the top of the bottom half.

The concept of an interchange is more easily understood in larger score groups than in smaller ones. In a 16-player group, for example, switching player 8 with player 9 looks like small potatoes. Player 8 would hardly have the right to claim that his measly 10-point rating advantage entitles him to an automatic pairing against number 16 while his virtually identically rated counterpart, player 9, must face number 1.

Of course, interchanges violate the sacred “top half vs bottom half” rule, so player 8 in the above example might be motivated to cite this perceived sacrament as though it were the only rule in the universe. He is quick to overlook that interchanges often produce best pairings for the tournament as a whole.

An interchange generally results in half-vs-half “violations” in two pairings. With only four players in the group, there ARE only two pairings, so it’s easy to overlook the validity of the concept.

In the case in question, if it were not for the presence of player G in the 3.5 score group, I would fully agree with the Swis-Sys pairings. A 10-point interchange, even though it results in a 330-point surprise switch in one of the players’ opponents, is preferable to a color history of WBWBB for one player and BWBWW for another. Finishing the tournament with two consecutive blacks is just horrible – much worse, for example, than BBWBW or WBBWB.

The existence of player G should, however, have saved the day. Dropping F instead of G into the next score group apparently would have solved all of the problems. Some TDs (and, apparently, some pairing programs) are excessively reluctant to transpose across score groups.

Bill Smythe

4

Just a thought about what SwissSys might have been looking at… it’s not JUST a matter of transposing F and G… as that leaves:

C, 3.5, 2130, BWBW vs. E, 3.5, 1930, BWBW
D, 3.5, 1940, WBWB vs. G, 3.5, 1890, WBWB
F, 3.5, 1900, WBWB vs. someone in the next score group

…which still leaves the colors wrong.

So it would then have to transpose E and G to fix the colors (still legal), to give what I’m assuming are more proper pairings of:

G, 3.5, 1890, WBWB vs. C, 3.5, 2130, BWBW
D, 3.5, 1940, WBWB vs. E, 3.5, 1930, BWBW
F, 3.5, 1900, WBWB vs. someone in the next score group

Now what I’m wondering is: does SwissSys see that one transposition won’t do it (it takes two), and thus considers one interchange to be better?

5

Quite possibly. By dropping F, you are essentially making two transpositions which total 50 points. The pairing the computer made required only one exchange, of 10 points. Under the current rules, the total rating difference is more important than top/bottom or transposition/interchange. There has always been a problem, however, determining how to apply these rules when a player is being dropped.

6

As player D in the above example, I was upset at having to play up 200 points vs down 40-60 points. Now I was due white last round, but I would have happily given up that “privelege” (please give me two blacks) if that meant i could have gotten my “due opponent,” the ~1900.

Especially since it’s a odd-numbered final round, adjusting for color alternation over top/bottom (even if an interchange) to me makes no philosophical sense. If that is a “correct” pairing, I think the rules should be changed.

Given prior color equalization, I have always been a fan of last-round coin flips myself. That way, “normal” pairings are not likely to be eschewed for mere alteration.

Ben Bentrup

7

It’s about time players, including Ben Bentrup, got over the notion that, just because they are in the top half of their score group by a few rating points, they have a God-given right to a lower-rated opponent in the next round.

All pairings are subject to transpositions of up to 80 points – including transpositions across the median line of the score group. Therefore, if you are in the top half, but within 80 points of someone in the bottom half, or vice versa, your pairing could go EITHER way. Learn to accept this and kwitcherbellyaching.

As for last-round colors, the mere fact that colors are equal so far isn’t the only important consideration. Having the same color in the final two rounds is a significant advantage or disadvantage, as the final rounds are those which count most, and in which colors are likely to be the most important. It is definitely worthwhile to make transpositions to avoid ending the tournament with two consecutive whites or two consecutive blacks.

Bill Smythe

8

Is there a test tournament available that one could use to test the pairings generated by various software packages? The test would include a list of players and their ratings and what the pairings should be for the next round. It would also include the results from the previous round.

Steve

9

That’s an interesting idea, I don’t know if Rules and/or TDCC would be willling to do it, though. I need to contact Tim Just on another matter, I may ask him about this, too.

The problem is be that a pairing program may be able to handle the test cases in isolation, but when two or three situations come up independently in the last round or two of a large tournament (say 150 players and 6 rounds) picking the ‘best’ pairing is a lot harder.

10

I’m not totally familiar with everyone here and I don’t know who Tim is. I’m not sure how difficult it would be to build a test tournament. It kinda seems that you could go through the rules and generate a result set that would test each one. Possibly you may need a couple of tests comprising a “validation” suite. And, I am not implying that this would be easy. But, as you said, it might be something worth looking into.

Steve

11

Tim Just, author of the 5th-ed rulebook.

The problem with creating a test-tournament is that the odd cases can only be found by experiment. Quick example: Last round, one player with 6, one player with 5.5 (has played the 6), two players with 5 (lower ranked player has played the 5.5), 8 players with 4.5. What’s the correct pairing? Sorry, time’s up.

Probably a more useful approach would be to supply student TDs with a real tournament (a U.S. Open, for example), and let them compare their pairing with the program’s

12

Better yet, take a real tournament and change a handful of results, just to create the awkward situations we all know and love.

13

Using a real tournament would be fine, but the pairings that were produced for it would still have to be validated against the rules. The assumption here is that the pairings for the tournament were produced with a software package but was that software package ever “validated”.

I think you could start with a real tournament but after the first round I think you would still have to maniupulate the results to test each part of the rule book. Here I am mainly concerned with Swiss Pairing.

Steve

14

It is really difficult to create a test event for pairings programs. Why? As Tom Doan, author of WinTD, pointed out to me that programs can make many legal pairings based on the rules; however, deciding on “best” pairings is harder. “Best” has not been defined and can be the subject of some debate even among good TDs.

One example comes to mind: When I was a director at a past U.S. Masters I entered the results and made the pairings with two different programs (SwsSys and WinTD). An NTD (not me) made a third set of pairings. Most rounds saw 3 different, yet, legal pairings in critical situations–sigh?!.

Tim Just
Chair TDCC
Editor 5th edition Rules of Chess

15

Tim,

Thanks for responding.

“Best” I guess would be a little evasive but is the result that multiple valid pairings can be generated a function of implementing the rules in a different order? I would think that given the same rule order, each program should produce the same result. Of course it is also dependent on ordering results e.g. equal ratings are then sorted by lastname, firstname, USCF ID, or something like that.

Steve

16

You are assuming that the rules, written in ordinary language, are directly convertible to a series of if-then algorithms. This is never going to be the case, although each edition of the Rulebook resolves a few more problems.

17

No, that’s not what I am assuming. What I am saying is that if each program evaluated the rules in the same order (not necessarily the order as presented in the rule book), then the results should be the same. If one program can evaluate the rules in some order and another program evaluate the rules in some other order, than perhaps one of those orders could be selected as the preferred order for evaluating the rules. In which case, all programs should generate the same results.

Steve

18

If you would actually read the rulebook, you would discover that, in many cases, the rules aren’t exactly intended to form an ordered list. Rather, they form a bunch (in no particular order) of various considerations to be taken into account when making pairings. How much weight to give each factor may vary from TD to TD, or from program to program, or even from circumstance to circumstance. This allows room for judgment, which can be EXTREMELY important, at least in the case of the most knowledgeable TDs. Don’t forget that TDs are allowed to overrule pairings made by the computer.

Bill Smythe

19

Steve,

Now you understand. The powers that be simply have not yet agreed on the order that rules get applied because they each see one rule as more important than another. Variations exist within the rules to allow everyone their own road to Rome.

Example: 29E6 addresses how to adjust pairings to improve colors. 29E6a is the “look ahead” method; however, 29E6b is the variation “top down” method. While both methods are legal they are different. They both have their champions with all sorts of data to back them up.

Tim Just

20

Tim,

Yes, the rules do allow for interpretation as to what rule is applied. But on the other hand, the FIDE Dubov system is a very precise method of doing Swiss pairings that when implemented provides for a deterministic outcome. It has some shortcomings, e.g. three of the same color in a row is not allowed, which allows ties when they could be eliminated by allowing the same color three times in a row.

I know the rules were developed to be very flexible, and they are, but perhaps too flexible. No criticism intended. In light of that, how are software programs validated for USCF pairings? Does the program get submitted to the USCF for review or does someone in the USCF work with the programmers or what?

Thanks.

Steve