[slyall]

There are pretty strong rules about how the draw on FIDE tournaments can be done and the programs that make them have to be certified. In one forum someone said they actually checked:

----

I see a lot of remarks saying "well, let them show the machine, so someone can check it", "It must be manipulated, this is so unlikely", etc.

The FIDE Swiss Dutch rules are on the FIDE website, in the handbook. There is pairings.fide.com which has a list of endorsed pairing software, meaning it was tested by FIDE to follow those rules. Why is nobody doing the checks?

Guess what? I did :)

Took the SwissManager tournament file from chess-results.com, created a TRF / FIDE rating report file, imported it, verified the pairings.

  Round 1: differences, which is to be expected: 
    people show up late, ratings get corrected, mistakes fixed, etc.
  Round 2, 3, 4: equal to the pairing in Gibraltar
  Round 5: a few differences in the group of people with 1.5 and 1 out of 4, nowhere near Hou.
    My educated guess: results of previous rounds were corrected after round 5 was paired
  Round 6, 7, 8: equal to pairing in Gibraltar
  Round 9: in the lower echelons 2 pairings were adjusted (the black players exchanged),
    due to (probably) Israeli not playing Iranian
  Round 10: equal to pairing in Gibraltar.

Does that count as sticking to the facts?

[maxander]

It could be interesting to check against a set of random seeds or random player lists. It's not entirely implausible that the algorithm is "correct," but that implementation details not covered by the spec mean that, e.g., gender imbalances in pairings occur. It might, for example, generate the pairings starting from a sorted list of players where an element of the sort key is gender.

[stonesixone]

If people agree that the software was correct but that the pairings were poor, there is always the possibility of reevaluating the algorithm itself. For example, an algorithm could be chosen that results in more "mixing."

[dsacco]

You can only optimize an algorithm in so many dimensions. Without making this a debate about affirmative action, I'd like to point out that if you're designing an algorithm to optimize for raw skill comparison in tournament match-ups, optimizing it to rebalance match-up results to be more mixed essentially voids that first optimization. In the aggregate, you're producing very different results by doing that. To put it simply, would you rather have skill-based fairness in a tournament or gender-based fairness?

You can design the algorithm in such a way that your priors are mistaken, or bias creeps in (though theoretically that can be self-corrected). But assuming that's not the case, and the algorithm correctly matches up mostly women v. women and men v. men in a skill parity optimization, you have a fair result for the purpose of a tournemant; i.e. skill-based match-ups.

If at that point you have an issue with the match ups for reasons of gender or ethnic parity, I would argue that you should seek to correct the upstream issues, not the algorithm. In other words, try to get more women playing chess - make the sport appeal to them more, make it more inclusive, etc. Rebalancing an algorithm is, in my view, a handicap, whether it's applied to gender disparity or any other disparity. I feel it does a disservice to both parties and doesn't really solve the root issue.

[logicallee]

Your argument contains a logical error!

>I'd like to point out that if you're designing an algorithm to optimize for raw skill comparison in tournament match-ups, optimizing it to rebalance match-up results to be more mixed essentially voids that first optimization

This is not necessarily the case. For example, if you alphabetized by last name, then obviously people with coincidentally the same last name could appear in any order. But if in twenty cases the men always were listed first (that's what the algorithm spat out), it might seem unfair. You could add the first name (another dimension) but you could also add a preference for mixing. Indeed, perhaps adding first names makes it unfair, as the pool of male first names is more skewed toward men (in the way aaron does not have a female equivalent). Last names likely have no such skew since a person born xx or xy gets the same last name.

So this example shows that the first dimension, which is fair (alphabetical by last name) can remain optimized while adding a second dimension. Because the first dimension doesn't care about what order people with the same last name appear.

Likewise perhaps the first dimension is equally fine with a few different pairings - so at that point optimize the second dimension.

[rtpg]

I don't know how Chess is classified, but isn't it possible that the men's side of the tourney just starts with more points due to historical imbalances?

For example, if we assumed that the current state had men having twice as many points as women of equivalent skill, it might take a while for people to climb up right?

More generally, it's not like people's skill are well ordered. If so, what would be the point of the tournaments! It seems like having a bit of mixing of levels in the beginning of the tournament would be more interesting. Especially if it's not single elimination

[stonesixone]

There can be a balance. One could introduce more mixing while still relying primarily on the current metrics. It's not a binary choice between one or the other.

[kaffeemitsahne]

It's actually a binary choice between a gender-biased algorithm and a gender-neutral algorithm. A small bias is still a bias.

[totalZero]

I don't fundamentally disagree with you, but I note that when you say "small bias," you're acknowledging that there is a continuum between total neutrality and total bias.

[4ifgnapoian]

I don't see how you arrived at the conclusion that the cardinality of biases is greater than the cardinality of natural numbers from the proposition(s) of the post you replied to. If that was the intent of your post, please elaborate.

[kakarot]

When people think they want a random output, usually what they really want is a pseudo-random output. People think randomness in some contexts is unfair because it isn't predictable. I don't think we should introduce gender-balancing to the algo, even though I doubt it's random to begin with. Once we cross that line, why not race balancing as well? People just need to be satisfied with the luck of the draw and redefine what "fair" means when dealing with things like random sorting or filtering.

[ianai]

Without those modifications does she wind up playing against fewer women?