If the 1–2 pairing is undesirable, then instead of moving 2 into the top group and doing a switch, could you just move 3 into the top group and pair 1–3, 2–6, 4–7, 5–8?
Yes, and that’s precisely my point.
The pairing above, even though it is by far the most logical one, would be rejected by WinTD (if Jeff Wiewel’s conjecture about WinTD is correct) because, by following Inconsistent Rule 2 rather than Inconsistent Rule 1, WinTD finds that pairing 1 vs 7 results in only a 1-point rating swap.
Inconsistent Rule 1 is 29D1b, which states that, in transpositions across score groups, the only rating difference that should be considered is the one involving players in the same score group. In our example, this means that pairing 1 vs 7 – i.e. starting with the raw pairings and then swapping 2 with 7 – would not even be considered, even though 1 and 4 (the raw opponents of 2 and 7) are so close together.
Inconsistent Rule 2 is 29E5c, which states that all transpositions should be evaluated based on the smaller of the two rating differences.
I sincerely hope that the latter was an oversight, and that 29E5c was actually intended only for intra-group transpositions. But, apparently, there are those who take 29E5c at its (literal) word, and end up with absurd pairings as a result.
Bill Smythe
Edited to change “inter-group” to “intra-group”. My apologies for the error.
The confusing question to me is whether the odd player who is dropped down from the higher group actually becomes a member of the lower score group (for purposes of the pairing rules), or whether lowest rated player in upper group vs. highest rated in the next lower group is a special case, treated separately by 29D. Transposition and interchange are defined only in terms of one score group. If the odd player and his opponent are not considered part of the score group, then 29E5C would not apply, and any switches to that pairing would fall only under 29D1b.
I never think of the dropped player as becoming a member of the lower group, nor of the raised player as becoming a member of the higher group. It’s a little better to think of this cross-score pairing as being a separate, two-player “group”. If one player has 2 points and the other has 1.5, I suppose you could think of it as the 1.75 group.
Even in this tiny “group”, however, there can still be transpositions. If the lowest 2.0 is paired against a 1.5 other than the highest, then, in effect, the highest 1.5 is being swapped with this other 1.5. It is precisely this situation where the transposition should be evaluated by looking only at the rating difference between these two 1.5’s, and not at the rating difference between their opponents.
Same applies, symmetrically, if a 2.0 other than the lowest is paired against the highest 1.5. Only the difference between this 2.0 and the lowest 2.0 should be taken into account.
I suppose that, if both transpositions are being made, i.e. if a 2.0 other than the lowest is paired against a 1.5 other than the highest, then both rating differences should be considered, but with an AND instead of the usual OR. In other words, the rating difference between the two 2.0s and that between the two 1.5s should be not greater than 200 points for equalization or 80 points for alternation.
Bill Smythe
I completely agree with this. The difficulty is that the written rules don’t say this explicitly, and sometimes seem to give contradictory guidance on how to treat this player.
In cases like this, I take the point of view that the contradictory written rules cancel each other out and give me license to do whatever I $%^&* well please.
Your milage may vary.
This is as good a point as any to make my semi-annual point that USCF pairing rules started going downhill when someone decided (about 3 editions ago) to try to write a computer-like program for humans to follow. The problem is that the programmers were using a very limited model of what could be done, and (as it turns out) weren’t very good programmers. In my view, it’s best for the pairing rules to lay out policy in the form of criteria (don’t pair the same players twice, equalizing colors is good, alternating colors is good, etc.) and let the human (or computer) do the best job they can of “maximizing goodness”.
Examples, case studies, and general guidance on how to proceed in pairing are all good - but they should not be taken as gospel, especially in hard cases. The saying is “hard cases make bad law”. It works the other way, too: “bad law creates hard cases”.
Computers alter the kinds of things you need to worry about. When humans make pairings, it’s at least possible that the TD is out to screw a particular player. Most players will accept that computers screw all the players equally.
More important - computer programs, running on modern computers, can explore a much wider variety of possible solutions to the pairing problem than humans can. Guidelines for “how to arrive at reasonable pairings” that work well for humans may actually prevent a computer program from coming up with far superior pairings.
In my view, the whole concept of “transpositions, etc.” is slightly flawed. The “natural pairings” are simply an expression of one of the criteria for good pairings. When humans do it, it’s almost always the place to start, and all changes from these pairings must be justified somehow. When a computer program does it, it’s not (necessarily) the case.
In my experience, the worst pairings I’ve ever seen are those made by early computer programs which tried to follow the printed procedures, without the experienced human looking over the shoulder to say “well, of course, in this situation that rule is silly - we don’t really follow that rule IN THIS CASE”.
But, over the past 5 years or so, I have never seen a pairing made by one of the top programs that I would consider it correct to change (unless, of course, the pairing TD had the settings wrong - in that case I would fix the settings and give the computer program another shot at it). I have listened to at least 20 very loud and contentious “discussions” among TDs - often 3 or 4 NTDs in the same discussion - trying to figure out whether or not to change the computer pairings. In EVERY ONE of these cases it turned out that: a) the pairing was obviously not “bad enough” to change, even if one side or the other was correct about the “best” pairing, and b) the TDs were missing a subtle point that the program had considered. Usually all of the pairings being discussed were “legal” and “defensible” and (in my opinion) would be upheld under any appeal - but my estimate is that in 95% of the cases, the computer pairing was correct, and the aging NTD kibitzers were defending past practices that are no longer BEST practice.
No need for a vote; 1-2, 3-4 good colors. If there is a good color option, take it.
Many people have reached the point where they feel that, in endeavors in various fields, a computer-generated solution is actually superior to a manually generated one.
I can’t remember which TD did the experiment, but one TD got tired of players complaining about his manual pairings. Before one tournament he got a pairing program, did the pairings manually, entered them into the program as manual pairing overrides, didn’t tell the players what he had done, and then heard the players comment on how much better the computer did the pairings. Using the computer program simply as an elaborate pairing typewriter for that tournament had the placebo effect of having the players more readily accept the pairings.
Not my experience. The programs work fine in “normal” pairing situations, but I’ve encountered several cases, generally involving odd color histories and people who had played different numbers of games, in which I had to override the computer pairing.
There is one bug to watch out for in some versions of WinTD (within the past five years). If the two middle players of a scoregroup have the same rating then the program is confused about where to split the group, and ends up (erroneously) pairing those two with each other and then the rest of the scoregroup normally.
Other than that I’ve found that all incorrect pairings are actually due to incorrect settings (and can be fixed by simply changing the settings and repairing - such as what was done for the final round for the top section’s top score group of the Illinois Open a few years back), or aren’t actually incorrect (I remember one Chicago Open I stopped by as a spectator while one player was complaining about getting the wrong due color - maybe 15 minutes before the round was scheduled to start - where the TDs were busy with other things and accepted my offer to look at what was done - it turned out that the player thought that he should get his due color based on higher ranking without realizing that full color history meant that his round five bye versus his opponent’s round five black was the overriding factor in the decision that rendered moot his round four black versus his opponent’s round four bye).
I have occasionally changed perfectly legal (and mathematically best) computer pairings, but generally only for things such as avoiding the pairings of family members (in a purely individual event) when there were reasonable transpositions available (I think it was a 12-point transposition when I last did this a few years back). When two family members competing for different class prizes are playing in the final round you risk ending up with the worst of all worlds. A) the family members may be upset about paying to play a tournament game against each other, potentially ruining one family member’s chances for a class prize. B) if one player wins then the players in the winner’s class may have questions about the game being ceded to give the winner a better shot at class prizes. C) if the game is a draw then players in both classes may have questions about whether or not a '“real” game was played or whether they were unfairly denied a chance to pick up a full point on the player in their class.
This sounds ideal, but in order to have any sort of consistency between TDs and/or computer programs, wouldn’t the goodness of each criterion have to be strictly quantified?
I wholeheartedly agree with this. But as a practical matter, limiting pairing programs to human-friendly algorithms usually guarantees that an experienced TD can understand and validate the computer pairings and explain them to players. By basing the rules solely on a prioritized set of criteria, you are bound to get into more situations where the computer pairings are not at all obvious to players, and a few situations where they can’t be justified by even an experienced TD without a detailed analysis. I don’t think this is necessarily a bad thing - on the contrary, it’s the price of progress - but it would require both players and TD’s to have more faith and reliance on computer pairings than they may have been used to.
Could we please see a few examples?