In “typical” transpositions NOT involving floats, it seems wrong (and completely contrary to USCF rules) to sum the two differences. The rule is clear – only the smaller of the two differences should count.
Now when it comes to floats, if you want to downfloat the highest instead of the lowest, the only relevant difference should be the difference between the highest and lowest.
Likewise, in upfloating the lowest instead of the highest, the only relevant difference should, again, be the difference between these two players.
So, if both are being done, I can see why both of these differences might be important, but the sum? That seems absurd. At most, it might be reasonable to use the greater (rather than the smaller) of the two differences.
There was a thread a few years ago, in which a pairing program (I don’t remember which one) was forced to make a 250-point transposition because two players had already played each other. Then, having made this transposition, the program then refused to make an additional 1-point transposition to equalize colors, apparently on the grounds that the 200-point limit for equalizing this player had already been exceeded.
That’s the kind of thing that can happen when rules, written in a general way rather than an algorithmic way, are converted into algorithms in pairing programs.
And I’m not sure what a good solution would be. Maybe something along the following lines:
- First make the “raw” pairings. Raw pairings are easily defined: Within a score group, top half plays bottom half, in sequence. If there is an odd number in a group, bottom player in group A plays top player in group B. Ignore colors. Also ignore whether two players have already played.
- In each “raw” pairing, assign colors in all the usual ways, especially the paragraph Pairing players due the same color in rule 29E4.
Sometimes raw pairings are called “natural” pairings, but I dislike “natural” because of its emotional content and biased nature, and because it’s hardly “natural” to pair the same players twice.
Then there is the next layer up, which I’ll call “simmered” pairings (or “almost raw”, or “barely cooked”), where transpositions are made only to avoid repeat pairings, still not taking colors into account.
The trouble with simmered pairings is that there is often more than one reasonable way to simmer the pairings. And, if simmered pairings are used as a stepping stone to the final pairings, the “best” simmered pairings may not lead to the best final pairings.
So, in evaluating any proposed set of final pairings, the proposed pairings should be compared to the raw pairings (not to the simmered pairings). Assign to each pairing (both raw and proposed) an “undesirability score”, something like this:
- If either player has the “wrong” color, this pairing gets 200 undesirability points for a bad equalization, or 80 points for a bad alternation.
- If the two players have unequal scores, this pairing gets 1000 undesirability points for each half-point difference in the players’ scores.
- If the two players have already played each other, this pairing gets 10000 undesirability points.
- If this pairing was different from the raw pairing, this pairing gets undesirability points equal to the transposition value. The transposition value of a pairing is defined as the rating difference between white’s raw opponent and white’s proposed opponent, or the rating difference between black’s raw opponent and black’s proposed opponent, whichever is less.
- The total undesirability score for a pairing is simply the sum of the above four. The total undesirability score for the whole set of pairings is just the sum of the undesirability scores for each pairing.
Note: The 1000 points for cross-score pairings prevents a program (or a TD) from wantonly making cross-score pairings just to improve colors. Likewise, the 10000 points for a repeat pairing prevents a program, in most cases, from making a repeat pairing.
By calculating the undesirability scores of both the raw pairings and the proposed pairings, the two sets of pairings can be compared. More importantly, two (or more) sets of proposed pairings can be compared. The pairings with the lowest total undesirability score constitute the preferred pairings.
The above method is designed to be compatible with USCF rules, including the 80- and 200-point color transposition limits. I’m not sure it actually is completely compatible, but complete compatibility between any algorithm and USCF rules may not even be possible.
Bill Smythe