Proposed Replacement for Rule 28 (Accelerated Pairings)

Running Chess Tournaments
1

This is a proposed replacement for Rule 28 on Accelerated Pairings. The main point of this is to replace the greatly flawed description of the 28R2 variation with a description which matches how (something like) this is implemented in WinTD. So far as I know, SwisSys only does added score acceleration, so this would just be ratifying the only known implementation of it.

The entire first paragraph seems more like an extended TD Tip than an actual “rule” and I’ve maintained that attitude in the redraft, which has pretty much the same information, just slightly rearranged to make clearer the description of how accelerated pairings work (and might fail to work, which seems to escape even some fairly experienced TD’s).

I propose renaming the old “Adjusted Rating” method as “Non-Winners” method, since I’ve eliminated the rather arbitrary rating adjustment, but if someone has a better idea for that, I’m certainly open to it.

The “rule” in 28R1 is actually just the first paragraph, which is the same as before, except for fixing a dangling “if any” in the handling of the odd player. I rewrote the expanded description to avoid mentioning “eighths” (the seventh eighth playing the bottom eighth from the old description is an “eyes glaze over” type of wording). I also removed the final (erroneous) statement “This method decreases the number of perfect scores”, since it might not.

The rulebook description of 28R2 looks like something that someone threw out there but never actually used. Despite taking up 1.5 pages, it has a lot of holes, incorrect claims and just some flat-out weird clauses. What this proposes is shorter, simpler, and more effective. Note well that even this does not achieve the rather outlandish claim from the rulebook that “This variation of accelerated pairings produces only about half the number of perfect scores achieved with the basic system. It therefore decreases the likelihood of multiple perfect scores, and causes the final standings to be more dependent on games between the top-rated players.” The difference vs added score is that it uses the top half draws as opponents for the bottom half winners, but it doesn’t change the number of bottom half winners, just provides slightly higher rated opponents for them.

How is this different from the version in the current rule book?

  1. It eliminates the 100 point temporary rating adjustments for players who drew (rather than losing) in round one. Given that the difference in post tournament rating for a player who has an extra 0.5 point is typically in the low teens, a 100 point adjustment is absurd and any type of adjustment is an unnecessary complication.
  2. The inclusion of draws (not just winners) in C2 is just flat out wrong. The whole flipping point is to try to give losses to the bottom half winners. The bottom half draws are already not perfect scores, so why mix them in?
  3. D2 (non-winners not paired with bottom half winners) are just paired normally (0.5’s with 0.5’s, 0’s with 0’s) since there is no compelling reason to treat them any other way.

The old language here for 28R3 is actually much more correct than the old language for 28R2. The only change is to the very last clause replacing “the non-winners of 5 vs. 6 against each other” with “the non-winners of 5 vs. 6 normally”

This proposes adding a section on accelerating beyond two rounds, which people do seem to want to do even if it doesn’t really accomplish much.

2

Just to reinforce the need for a change, I’ve explained to multiple NTDs that questioned me opting for the +1 method (they wanted to use the adjusted rating method) that the current rule can have a round two with upper half 1/2-1/2 players paired against lower half 0-1 players (top 16 players have 4 decisive games and 4 draws, giving 12 non-winners in B2 while the bottom 16 players have all decisive games with the top 8 bottom-half winners in C2 playing the top 8 top half non-winners while the bottom 4 top half non-winners (possibly including the number 16 person that drew) paired in the D2 group with the 8 bottom half losers. So player 30 could lose to player 22 in round one and then get paired in round two against player 14 that drew in round 1 against player 6 (while 15 and 16 had upset wins over 7 and 8 ).
Any time the top half has more round one draws than the bottom half, the rule as currently written would have some bottom half 0-1 players paired against a top half player in round 2.

3

If you really want to tick off players, announce before the first round that accelerated pairings will be used. Players realize that they will not be able to figure out pairings before round 3. In that round they expect to see mismatches that normally arise in round 1. 28R is a reflection of poor planning on the part of an organizer. Instead of having a multiple section tournament or a class section event, he prefers to hold a one section event because it appears to be easier and cheaper to run. In the days when one section tournaments were the thing to do, all sorts of messy options were tried to come up with one winner, thus the mishmashed guidance of 28R. The assumption was that high rated players ultimately wanted to play other high rated players. That wasn’t true, as high rated players wanted to win money and did not care if they did not face a rival. Any method that would kill off rivals and dangerous up and coming young players was fine with them. They did not fear multiple ties for first if it meant money in the pocket. If it was a two day event, a high rated player would often get a mismatch in round 3, which meant he would get an easier win and a chance to go home early. After many complaints about how unfair accelerated pairing were to the middle and low rated players who were trying to win class prizes, organizers started to have multiple section tournaments to appease their customer base.

Most of the accelerated pairing methods were used in two day, 5 round events. After going 1-1, rather than the normal expectation of 2-0, some higher rated players would withdraw and not come back to play in such tournaments. That was bad for business in general. 4 round tournaments usually did not have accelerated pairings because the organizer was smart enough to have two or more sections based on rating. The TD was happier, too, as he did not have the time or desire to explain how accelerated pairings worked. In scholastic tournaments, people complain that accelerated pairings mess up tiebreaks. Whether or not that is true, the perception of that occurring makes some scholastic organizers reluctant to even think of applying 28R. Maybe the accelerated pairing rules should be abolished. It would not be missed except for rules mavens who like to tweak and argue about everything.

4

When accelerated pairings were first made available they were looked upon as a way to get better pairings with fewer mismatches, but it only really works that way for the top and bottom eighth at best. They do reduce the number of perfect scores and thus are good when the number of players in a section well exceeds 2 to the power of the number of rounds. Accelerating without using the +1 option can make for some very ugly pairings on the top boards in round three and the bottom boards in round two (well, at least if you follow the the rulebook verbatim instead of Tom’s variation) . Heck, even accelerating with the +1 option can still on occasion make for such ugly pairings (in non-US-chess-rated HS team-vs-team four-round events (more than 40 teams - splitting it into two sections would risk the state HS association not counting it as a valid tournament for qualifying for state) I’ve taken the two lower half 2-0 teams and paired them against the strongest 1-1 teams while the remaining four 2-0 teams, all strong, have natural pairings for the final two rounds, getting a reasonable reduction to one perfect score - not an option in US Chess rated events unless using it as a publicized major variation).

Nowadays I most commonly see accelerated pairings in scholastic trophy tournaments where you want to reduce the number of perfect scores and where team standings considerations would generate a strong push-back if you tried splitting a grade-based section (very strong push-back from the teams that would get split in half).

Using Modified Median before Solkoff reduces the concerns about tie-breaks and most of the people I’ve dealt with (local and national) will accept that muddling in exchange for the better chance of not using tie-breaks to award a championship trophy between multiple players with perfect scores.

Because of the possibility of a 32-player four round event having a 0-1 player 30 getting paired in the second round paired against a 1/2-1/2 player 14, I’ve always used the +1 option and that is a lot easier to explain to people. My gut feeling is that the +1 option used appropriately is similar to adding around two thirds of a round to the schedule for reducing perfect scores, while the rating adjustment version per the current rulebook is closer to adding one third of a round.

5

There are situations where it is necessary to have one section. Even if other sections are added, many people may choose to play in “key” section. State open championships, scholastic championships, combined team and open events are all examples.

6

To avoid the possibility of multiple round 2 pairings of a top-half 0.5-0.5 vs a bottom half 0-1 I would recommend the following.

That avoids the strangeness of a 5-round 64 players section with A1 vs B1 having 8 decisive games (higher rated player winning) and 8 draws (boards 5 through 12), C1 vs D1 having all wins by C1, and round two dropping 0-1 players 29 through 32 to be paired against 0-1 players 51 through 54 after those players had already lost to 33 through 36.

Whenever round one has more draws in the top half than the bottom half then B2 will be larger than C2 (well, barring round two byes more heavily distributed amongst top half players). One option for B2 vs C2 is to treat the bottom-ranked (score then rating) B2 excess (an even number) as a lower score than the remainder of the B2/C2 combination (which will all be paired as having the same, higher, score) and pair the group normally.

7

No. The remainder is paired normally (which is also a change from the old rule, whose treatment of the extras made no sense). So, in your case, the extras in B2 that float down to D2 are four draws (25-28) and four losses (29-32). 25-28 would be paired among themselves, while the 0-1 score group in D2 would have 29-32 then 49-64 so the top half would be 29-32+49-54 which would play 55-64.

8

I corrected the second paragraph with 0.5-0.5 but not the first. Thanks for catching it.

I was trying to finish the post before going out for the day and thus made an error in the count. You are correct that in the second paragraph it is 55-58 (not 51-54) that would have lost to 39-42 in round one only to end up paired against 29-32 in round two.

With my suggested modification the B2/C2 group would have 5 through 12 and 17 through 28 paired against 29 through 48 (C2 plays the 9-12 and 17-28 instead of 5-12 and 17-24) while the 49-64 players at 0-1 would be paired against each other.

Then I continued the discussion by suggesting that maybe first the excess losses would be floated down but paired against each other (essentially a different group between C2 and D2), leaving C2 to be paired against the 0.5-0.5 players 5-12 and 21-28 with 17-20 playing 29-32 and 49-64 paired as its own group. In that case, the difference from your original version is that the bottom half of C2 faces players 21-28 that drew instead of 17-20 that lost and 21-24 that drew. If ratings are trustworthy enough to do accelerated pairings then 21-28 should still beat 41-48 and if ratings are not as trustworthy then the pairings would still include players who have proven they are strong enough to draw in the top half instead of players who have not done so.

Creating an additional group of only 0-1 top half players may be questionable. If those games are decisive then you still end up with 0-2 top half players (4 in my example) paired in round three against 0-2 bottom half players instead of having four or eight such pairings in round two with 0-1 players. If all the 0-1 top half players are kept in the B2/C2 group then there can still be 0-2 top half players against 0-2 bottom half players in round three (four in my example if the higher-rated player wins on boards 5-32).
With my round one example results and +1 pairings then 17-20, 29-32 and 33-36 would play 37-48 and there would be a good chance for round three having four 2-0 bottom half players paired against four 2-0 top half players (the notorious round three strange pairings when using +1). With the original revision proposal the odds for that go way down while retaining the cost of four 0-1 bottom half players playing four top half players in round two and now there being very little chance of the +1-version problem having multiple round three 0-2 top half vs 0-2 bottom half pairing. With either of my modifications round two looks better from the standpoint of the bottom half players but the +1-version round three problems are still there.

Using my example round one results it is an interesting question whether Tom’s original revision (C2’s 33-48 plays 5-12 and 17-24 with 17-20 having lost) or my modifications (either 33-48 playing 9-12 and 17-28 with 17-20 having lost or 33-48 playing 5-12 and 21-28 with no top-half losers) are significantly different in reducing the bottom half 2-0 results. Looking at the bottom-half losers, +1 pairings have no chance of round two having top-half vs bottom-half 0-1 pairings but a good chance for round three having four top half-bottom 0-2 pairings, either of my modifications retain four or two of those bad 0-2 pairings, and the original revision has very little chance of any of those bad round three pairings at the cost of round two having four top-half vs bottom-half 0-1 pairings.

My suggested modifications were proposed primarily to reduce the ugly top-half vs bottom-half round two pairings between zeroes that the current rule has. After taking another look at the originally proposed revision, those bad pairings remain in round two while my modifications would push them to round three (where they already are with +1 pairings). The reduction of bottom-half 2-0 players (compared to +1) happens with or without my suggested modifications, and is more likely if the top half has its excess players float down out of the B2/C2 group - kill my first suggested modification of keeping everybody in the B2/C2 group and instead have the C2 group plays the strongest players in the B2 group.

Taking another look at the creation of a floated-down group betwen B2/C2 and D2, the floated-down top-half 0-5-0-5 players would play each other and the floated down top half 0-1 players would play each other. That would end up with round three having two pairings with top-half vs bottom-half 0-2 players, which is half of what you might expect with +1 pairings. It is nice to reduce the number of gross mismatches but it is much, much better to retain the bulk of the acceleration strangeness in the first two rounds, so kill the second suggested modification of having a group between B2/C2 and D2.

The only question left is whether or not a bottom-half 1-0 player is more likely to beat a top-half 0-1 player (who may be over-rated or having a bad tournament or had one unfortunate lapse) or a somewhat lower-rated top-half 0.5-0.5 player (who may be under-rated or having a good tournament or have been lucky). Upon further review, I’m not sure whether dropping the somewhat higher rated 0-1 players or the somewhat lower rated 0.5-0.5 players is a better option, so the simplicity of dropping the bottom rated excess players regardless of score seems reasonable.

9

So far, so good. The above combines the three paragraphs of 28R into two, makes several improvements, and warns TDs a little more sternly than before about situations where accelerated pairings probably won’t work.

I think some TDs simply think, “Look at that, NN players. Let’s use accelerated pairings and cut the number of perfect scores”, without really understanding how accelerated pairings may accomplish that objective, and worse, why and when accelerated pairings may not work as intended and may even backfire.

Bill Smythe

10

Sounds good. You could make “(if any)” dangle even less by removing the comma after it, so that the first sentence would read “… the players are numbered, ordered by rank, byes removed, the odd player (if any) assigned a bye and removed, and …” :slight_smile:

This is a big improvement too.

Bill Smythe

11

It seems to me that the first half-sentence of your proposed 28R1 also applies to 28R2. So 28R2 could begin as follows:

"Before the first round, the players are numbered, ordered by rank, byes removed, the odd player (if any) assigned a bye and removed, just as in 28R1. Then the field is divided … " [etc]

This is followed by a list of procedures involving A1, B2, C3, D4, K9, WD40, etc, which is an improvement over the existing version. And then:

I’ll take your word for it. In any case, the artificial 100-point additions to, and subtractions from, the ratings of players who draw are eliminated, and the number of perfect scores is, in all probability, reduced more effectively than with the current version.

I question, however, whether any version of 28R2 will be needed at all now. Your suggested improvements in 28R1 seem to do the job just fine, rendering 28R2 pointless. (But maybe it should be kept, in order to mollify the Delegates.)

Bill Smythe

12

Remember, however, that 28R1 has the problem that if there are more draws in the top half than in the bottom half (which would probably be a common situation), then there will be more bottom half 1-0’s in the artificial 1 group than top half 0-1’s, so someone can get to 2-0 playing two players near the cut line for the bottom quarter.

13

Discussion of proposed rule changes on the forum can be A Good Thing™. However, the Rules Committee is trying to encourage a culture shift. We strongly urge proposed rule change ADMs (perhaps hashed out on the forums) to be sent to the Rules Committee, if possible allowing some time before the ADM deadline for the committee to review the proposal and possibly suggest refinements. (So, for instance, sending in an ADM on May 31 when the deadline is June 1 is more likely to get you a “good luck, pal” than a detailed discussion and review.)

The point is not to discourage members from suggesting rule changes. The fate of most ADMs that haven’t been vetted by the Rules Committee in advance is to automatically be referred to the Rules Committee for the next meeting. Working with the committee in advance will help move the ADM along.

14

I have never used accelerated pairings at any event that I’ve directed (and I’m not a big fan of them as a player), but I thought I understood the basic concept pretty well. Reading through this thread, though, it all seems impossibly complicated! If I ever want to use them, I’ll have to take a course or something (or at least do a detailed study of these posts).

15

+1 acceleration is fairly simple to explain and does a fairly good job of reducing perfect scores. The more complicated version (with the different way it handles draws) that Tom wants to change does a slightly better job of reducing the number of perfect scores and Tom is trying to make it do a definitely better job.

I’ve used +1 acceleration for a number of 4-round team vs team tournaments (sometimes accelerating by sixths with 40+ teams). There are very few tied matches because they used board-weighted scoring. Acceleration has been very effective in getting the field down to one 4-0 team. For those particular events both +1 and the more complicated version would have given the same pairings because of that rarity of tied matches. One time I did end up pairing two lower-rated 2-0 teams (that had scored an upset win in the second round) against the top 1-1 teams while pairing the four high-rated teams together so that I could reduce the field to one 4-0 team (it was not a US Chess rated event and the coaches all understood and accepted the variance from normal Swiss pairings).

One common mis-perception is that acceleration avoids big rating mismatches. I’ve had players in club tournaments request accelerated pairings (when they weren’t needed) to get “better” pairings. Once I showed them that the round three pairings in the middle of the field had larger rating differences under acceleration then the players would stop making that request.

If you want to use acceleration then it helps to understand that broadened middle.
Example - 64 players, five rounds, higher-rated player wins every game
Standard pairings have 1-32 beating 33-64 in round one and 1-16 and 33-48 winning round two against 17-32 and 49-64, leaving 1-16 at 2-0, 17-48 at 1-1 and 49-64 at 0-2 (17-32 play 33-48 in round three).
Acceleration (both versions are the same with no draws) has 1-16 and 33-48 beating 17-32 and 49-64 in round one, and 1-8, 17-32 and 49-56 beating 9-16, 33-48 and 57-64 in round two, leaving 1-8 at 2-0, 9-56 at 1-1 and 57-64 at 0-2 (9-32 play 33-56 in round three).
Things start settling down later and going into round five (with the higher rated always winning) the 2-2 group would be players 21-44 with standard and 19-46 with acceleration. 3-1 is 5-20 vs 3-18.

16

As a rule of thumb, if I don’t expect many draws between the stronger players then I will accelerate at 1.5 times 2 raised to the power of the number of rounds (for a five round tournament, two to the fifth is 32 and I will accelerate at 48 and only very rarely have two 5-0 scores). The trustworthiness of ratings is crucial to acceleration so I don’t accelerate the K-1 sections unless exceeding 2x, and even then I wouldn’t if the majority of the section is unrated.

If you do expect a lot of draws then you many not want to accelerate even at 2x instead of 1.5x (for the National scholastics the number has to exceed 2x before accelerating, which is means more than 256 for a 7-round tournament or 128 for a 6-round tournament). For the National JH and HS 7-round tournaments (with their higher percentage of upper board draws) I’ve seen acceleration get down to one 6-0 player with a round to go, and at least one time the JH was not accelerated even when exceeding 260 and it still had only one 7-0 player. The under sections of national scholastic are not accelerated and even their lower draw rate resulted in at most one perfect score in the bulk of the large sections at SN-VI.

17

What nobody is talking about is the idea of accelerated pairing being used for more than 2 rounds. I will admit that this is the standard in the US, and I have never heard of a US tournament accelerating more than 2 rounds. That can not be said for the rest of the world. A few years back I played in Sardinia and accelerated pairings [100 players with 9 rounds of play]. were used for the first 5 rounds. They used a system developed in France call “Degraded Accelerated Pairings”, and I think the name says it all about that crazy system. In the third round of the event a 2200 rated player with 2 losses was playing a 1600 rated player with 2 wins.

Larry S. Cohen

18

What makes this even more bizarre is that there is no reason to use accelerated pairings at all with 100 players and 9 rounds. With plain old normal pairings, there would be guaranteed to be no more than one perfect score after 7 rounds, let alone 9.

19

I’ve said in this thread that I’ve partially accelerated round three in non-US-Chess-rated events.

In the past, Bill Smythe has talked about the latter round acceleration done by Richard Verber in a scholastic Fischer-boom-era event.

20

If it’s 9 rounds, it may be some system devised to make it easier for higher finishers to earn FIDE norms. FIDE pairings aren’t allowed to be ad hoc (i.e. you can’t have player A play up rather than down in the last round to increase chances of a norm) but if you come up with a byzantine system of adjustments based solely upon scores and ratings (and not title possibilities) it would probably pass muster.

However, note that I did add an extra category for acceleration beyond round two.