why do my 'hand' pairings differ from SwisSys

after 2 rounds 14 players have a score of 1. The ‘natural pairings’ are:
2075 wb vs 1811 bye-w
2037 bw vs 1801 bw
1990 bw vs 1800 bw
1989 bw vs 1788 wb
1956 bw vs 1748 bw
1872 wb vs 1746 bw
1828 wb vs 1494(provisional) bye-b
But the ‘natural’ pairing would cause several players to have the same color twice in a row. So I left the top board:
2075 wbW vs 1811 bye-wB
then I transposed 1801 with 1800:
2037 bwB vs 1800 wbW
And now the next natural pairing, 1990 and 1801, have had the same color sequence so I transpose 1990 with 1989:
1989 bye-bW vs 1801 bwB
Now I can’t pair 1990 with 1788 or 1748 without repeating colors, so I transpose 1746 over both and get;
1990 bwB vs 1746 wbW
1956 wbW vs 1788 bye-wB
Up to this point SwsSys pairing matches mine.

Now there are four players left to pair: 1872 bw; and 1828 wb; and 1748 bw; and 1494 bye-b. 2 players are due white and 2 are due black.
Again, the natural pairing of 1872 with 1748 causes a repetition of colors so I transpose 1872 with 1828 and pair:
1872 bwB vs 1494 bye-bW
1828 wbW vs 1748 bwB and on every board everyone gets the alternate color.
BUT SwsSys paired the last four players 1872 bwB vs 1748 bwW and 1828 wbB vs 1494 bye-bW so that both 1748 and 1828 repeated colors.
I assume that SwsSys is the correct pairing, but I can not find in the rule book why my pairing is wrong and its is right. Can anyone shed some light on this for me?

Do you have the correct color histories above? I don’t understand your pairings.

Alex Relyea

Also, have any of the players met before?

Alex Relyea

The color histories of some players differ between the two sets of players.

More importantly, the opponent history is missing, as is any comment about things like state avoidance, team avoidance and club avoidance.

As others have commented, much more information would be needed before this can be analyzed. However, I would strongly discourage you from believing there is “the correct pairing” under US Chess rules. Sometimes, sections are straightforward to pair, and most TDs would agree on what constitutes “good pairings” for the section. Other times, however, there can be multiple ways to pair a section, all of which are “good” under US Chess rules.

There is a joke that you can give n NTDs a section to pair, and they will come up with n+1 ways to pair the section. :smiling_imp:

First of all, you made several typos in the colors you stated, either in your raw (“natural”) pairings at the top or in your narrative at the bottom. They don’t match. I’ll assume the colors given in your narrative are correct, and I’ll make the corresponding corrections in your raw pairings. The corrected colors (corrections bolded in blue) then seem to be:

2075 wb vs 1811 bye-w
2037 bw vs 1801 bw
1990 bw vs 1800 wb
1989 bye-b vs 1788 bye-w
1956 wb vs 1748 bw
1872 bw vs 1746 wb
1828 wb vs 1494 bye-b

Assuming the above version is correct, I see no reason not to make the pairings you suggested – unless 1872 has already played 1494, and/or 1828 has already played 1748.

Also, were any of these pairings not chosen because both players were from the same team, family, club, or school?

When you post a question like this, it would be helpful if you could give us a complete crosstable (at least within the score group being questioned), including colors, opponents, and round-by-round results.

Bill Smythe

I’m guessing here.
1746 to 1801 is only 55 points and is within the 80-point limit for transpositions to alternate colors even though 1872 is 165 points away from 2037. However, 1828 is more than 80 points from 1956 (and 1494 vs 1748 is more than 80 points) so it wouldn’t bring the 1748 all the way down to the bottom board just to alternate colors.

Personally, I’d figure that if there are two color conflicts with the raw pairings (one game with both due white and one game with both due black) and still the same two color conflicts with the adjusted pairings then there is no reason to make any changes at all just for colors.

Note that swapping the 1828 and 1801 fixes everything and is only a 27-point interchange (actually the three way-swap with the 1811 paired against the 1494). I’m guessing the pairing program setting was to avoid interchanges.

I can’t check right now, but I’m fairly certain SwissSys (unlike WinTD) does not have a setting to avoid interchanges.

In the OP’s (original poster’s) description of how he would make the pairings, I noticed some non-standard (though not entirely incorrect) thought processes in making the pairings by hand.

The first thing that should be done in hand-pairing a score group is to count the number of players due white and due black in each half of the score group. In this case there are 4 due white and 3 due black in the top half, and 3 due white and 4 due black in the bottom half. (Overall, 7 due white and 7 due black.) Therefore it should be possible, with any luck, to assign everybody the correct colors.

If, however, there had been (for example) 8 due white and 6 due black, you would know right away that it would be impossible to give everybody their due colors. In at least one pairing both players would necessarily be due white. In this case, there may be no point in trying to transpose the first “raw” pairing you find where both are due white. You would know ahead of time that such a transposition would accomplish nothing overall. If, on the other hand, you found a “raw” pairing where both are due black, you would definitely want to transpose if at all possible; otherwise, the color problem would be worsened and you would end up with three bad-color pairings instead of just one.

Another thing to consider (in some situations) is whether any player is due neither color – i.e. has had nothing but byes or unplayed games in previous rounds. If so, such a player should be treated as though he is due the color that would reduce the total color imbalance in the score group. If, for example, there are 7 players due white, 6 due black, and 1 due neither, the player due neither color should be treated as though he is due black. That way everybody (Lord willing and the creek don’t rise) can receive the color they are expecting.

The same concept applies if some players are due their color in order to equalize (e.g. bye-b), while others are due their color merely in order to alternate (e.g. wb). For example, if 8 are due white and 6 due black, then it would be better to give the “undue” black to wb rather than to bye-b.

All of this is part of the “look ahead” approach. If you simply use a “top down” approach (trying to transpose each bad color as you find it), you may end up making transpositions that accomplish nothing overall.

One thing I found peculiar about the OP’s pairing process was that he sometimes transposed players in the top half with each other. In general, it is better to make the transpositions in the bottom half only. Think of the top-half players as being in fixed order by rating, while the bottom-half players are moved around as necessary. (Nevertheless, when considering a transposition, look at the rating differences in both halves when deciding whether you have a 200-point violation or an 80-point violation.)

Using this philosophy, you would come up with the following pairings:

raw pairing 2075 wb vs 1811 bye-w: OK as is.
raw pairing 2017 bw vs 1801 bw: Colors don’t work, so instead look for the next bottom-half player due white. Thus:
revised pairing 2017 bw vs 1800 wb.
Continuing from there:
new raw pairing 1990 bw vs 1801 bw (1801 hasn’t been paired yet). Colors don’t work, and the first remaining bottom-half player due white is 1746 wb, so:
revised pairing 1990 bw vs 1746 wb.
The next two raw pairings work out OK:
new raw pairing 1989 bye-b vs 2017 bw.
new raw pairing 1956 wb vs 1788 bye-w.
The remaining new raw pairings are:
new raw pairing 1872 bw vs 1748 bw.
new raw pairing 1828 wb vs 1494 bye-b.
Colors don’t work, so these last two pairings are transposed:
revised pairing 1872 bw vs 1494 bye-b.
revised pairing 1828 wb vs 1748 bw.

Lo and behold – these pairings are exactly the same ones arrived at by using the questionable method of making transpositions in the top half. That won’t always be the case, however. I’d suggest always making the transpositions in the bottom half.

Other (older) threads that might be of interest are:

Pairing Question
Three-Way Transpositions
Point-Count Pairings?

Bill Smythe

Note that the full 1748 transposition (for color alternation) results in taking the place of the 1494 (way over the 80 point limit for alternation) and in playing an 1828 instead of a 1956 (more than half again over the 80 point limit for alternation).
The interchange with the 1828 taking the spot of the 1801 is only 27 points and affects only three pairings (two if you don’t do the three-way transposition).

When the color imbalances are opposite among the top half and bottom half then, if transpositions don’t clean things up, remember to look at interchanges.

In my opinion, the correct way to evaluate compound (e.g. three-way) transpositions is to look at each proposed pairing and, for each of the two players in that proposed pairing, consider the rating difference between that player’s raw opponent and his proposed opponent. If the lesser of the two differences is within the transposition limit, then the proposed pairing is acceptable.

If, for example, the raw pairings are

A 1900 vs D 1700
B 1840 vs E 1635
C 1800 vs F 1585

and one proposes (for colors, or to avoid rematches) the three-way transposition

A 1900 vs E 1635
B 1840 vs F 1585
C 1800 vs D 1700

then the three proposed pairings are evaluated as follows:

A vs E: The lesser of 1700 minus 1635 (the difference between A’s raw and proposed opponents) and 1900 minus 1840 (the difference between E’s raw and proposed opponents). Thus, the lesser of 65 and 60, which is 60.

B vs F: The lesser of 1635 minus 1585 (the difference between B’s raw and proposed opponents) and 1840 minus 1800 (the difference between F’s raw and proposed opponents). Thus, the lesser of 50 and 40, which is 40.

C vs D: The lesser of 1700 minus 1585 (the difference between C’s raw and proposed opponents) and 1900 minus 1800 (the difference between D’s raw and proposed opponents). Thus, the lesser of 115 and 100, which is 100.

The greatest of these three lessers is 100, which means that this proposed transposition would be acceptable in order to equalize colors (or, of course, to avoid rematches) but not merely to alternate colors.


In the actual case at hand, though, we have a six-way transposition. (The top proposed pairing is the same as the raw pairing, so only 12 of the 14 players are in the transposition.) This six-way transposition does not divide into two 3-way transpositions, nor three 2-way transpositions, nor one 4- and one 2-way transposition. To see this, place the top and bottom halves in parallel columns, and start with any player, going from left to right via the raw pairing and then right to left via the proposed pairing. You will find that you never get back to the starting point until all 12 players are included in the circle: 2L - 2R - 4L - 4R - 5L - 5R - 7L - 7R - 6L - 6R - 3L -3R - 2L. (Players are 2L through 7L in the left column, 2R through 7R in the right. Players 1L - 1R are not part of the transposition.)

If you apply the above-explained arithmetic to each of the six proposed pairings, you will find that the first five proposed pairings all evaluate to less than 80 points, while the sixth (7L - 5R) evaluates to 128 points. Thus, the six-way transposition is acceptable to equalize colors, but not merely to alternate colors.

So which is it, equalization or alternation? Some of the 12 players are due their colors in order to equalize, so it might be reasonable to argue that this entire six-way transposition is acceptable.

On the other hand, it might be equally reasonable to argue that, since both players (7L - 5R) in the sixth proposed pairing have alternation as their only issue, the transposition is not acceptable. Apparently Jeff holds this view.

So, bring on the interchanges!

Bill Smythe

You can fix all the issues other than the alternation on the bottom boards without going above 80. Fixing the last two requires going above 80. Ergo, you don’t do it. I don’t really see how one could argue (with a straight face) that if you fixed an equalization somewhere in the score group, you have carte blanche to do 200 point switches anywhere else.

Am I the only one who really wishes Mr. Mackey would follow up to his original post?

Alex Relyea

No.

No.

We still need the crosstable. Probably, several of the pairing sets suggested so far involve pairing the same opponents twice.

Bill Smythe

There is an interesting philosophy question about pairings for me. Should our pairing rules be sufficiently stringent and detailed to meet FIDE’s description of pairings. An example is the differences between the occasional pairing differences produced by SwissSys and WinTD. I am certain that Tom Doan will have interesting comments especially after the discussion of full point bye assignments in another thread.

An alternative option might be to have a stringent set of rules for computers but leave the current set of rules alone for hand pairings. This might have some problems when a TD or player disagrees with the computer pairings. Which set of rules then apply?

Regards, Ernie

Some modifiers:

  1. team avoidance (mandated in many scholastic tournaments)
  2. club avoidance (such as avoiding a father-son pairing in the final round with both in the running for prizes)
  3. state avoidance (done for the early rounds in some nationals)
  4. other avoidances (such as not pairing a player from Lilliput against one from Blefuscu when their countries are at war)
  5. varying transposition and interchange limits
  6. ignoring color-based transpositions (transposition and interchange limits set to zero) to account for double-round pairings (two games per round against one opponent with colors switched in the second game)

A pairing program (or person) ignoring some of the modifiers (or using some that shouldn’t be used) can come up with a different pairing from an entity using the correct modifiers. You can see identical versions of the same pairing program running on two different machines having the same hardware and software, but get two different sets of pairings because one program has the wrong modifiers.
A manually determined pairing may be the correct one in such circumstances eve when there are no special allowances for manual pairings.

Certainly, it would be desirable to have “deterministic” pairings. However, the Dutch system (and probably even more so the Dubov) are pairing systems that were (are) designed for the types of Swisses that are common for at least semi-professional chess players, not for the vast majority of tournaments that US Chess rates.

  • There’s an implicit assumption that players with identical ratings are rare. As pointed out in the thread on byes, the bye in the Swiss would go to the unrated player who is last in alphabetical order. The fact that the “tie break” in ordering players is something as arbitrary as alphabetical order is OK if only rarely needed.
  • The elevation of color over rating difference (not matter how extreme the latter is) is more in line with the preferences of top players than most US Swiss players. (It’s also not going to be common to even have massive rating differences in (semi-)professional tournaments.
  • Big score groups can be a problem because we often have things like state or team pairing preferences. The number of potential pairings in a score group of 2N players is N! even if you ignore interchanges. Even N=20 is prohibitive. However, if just about everyone in the top half can play just about everyone in the bottom half, which would be the normal situation if there were no optional preferences, the Dutch system can find that one “correct” pairing fairly easily. It’s when you have a substantial number of blocked pairings that the fix-bottom-up-with-backtracking can take too long to be practical.