Comments on Kaps Pairing System

Canadian Open Chess Championship,
Kapuskasing, Ontario, July 12-20 2003

by Jonathan Berry, International Arbiter

Skip the verbiage and cut to the chase! What are the exact pairing procedures for the 2003 Canadian Open? See synopsis. You might want to return here afterwards.

Introduction to Pairing Systems

Chess is a game between two players. Who plays whom and when is called Pairing.

The quintessential pairing system is the Match. If there are more than two players, you can have the Elimination or Wimbledon system, where only the winner of each match advances to the next round. Or you can have the Round-Robin or All-Play-All. The first relies on a careful tailoring of the field so that the number of players is 2 to the power R, where R is the number of rounds. The latter requires the number of players, if even, to be one greater than the number of playing sessions.

The Swiss System

If the number of players is flexible, and much greater than the number of playing sessions, you need a different approach. Enter the Swiss System. The systems discussed here all rely on the traditional scoring system of chess, where a draw is worth half of what a win is worth, and a loss is worth nothing. If you can live without that stricture, I recommend the Haida Pairing System.

In the Swiss System, pairings are made before each round. In general, a player meets another player with the same number of points. At the end, the standings are determined by who has the most points. Nobody is eliminated.

Sounds simple, but the devil is in the details.

How to Pair the Group - various approaches

A score group comprises the players who have the same number of points. Ideally, a group consists of an even number n players, numbered, not surprisingly 1, 2, 3, ..., n-1, n in order of rank, with 1 being the highest rank. Ignoring the special procedures which deal with colour allocation, here are the various approaches to pairing the group

• Top-versus-Top  TvT  1.2, 3.4, ..., n-1.n This has never had much following. On the plus side, players meet equals. On the minus, it does not sort the players by strength.
• Top-versus-Bottom  TvB  1.n, 2.n-1, ..., n/2.n/2+1 Used for a while in British Championships (1970s?) and recommended by me to FIDE in 1988 for elimination events. After using other methods which produced a lot of upsets (and thus marquee players missing from the later stages), FIDE did adopt it for their world championship. It gives the top players an easier ride and avoids the discontinuity at the middle in TvM pairings. It is perhaps best suited to play within a narrow range of strength where, for example, most of the field is separated by less than 200 points.
• Top-versus-Middle  TvM  1.n/2+1, 2.n/2+2, ..., n/2.n The method used in TSP, in FIDE and CFC pairings; most people consider this synonymous with The Swiss System.
• Random A popular method in the old days when players did not have ratings, or when comparison of various systems was thought to be unreliable, or when the use of ratings was thought to favour certain players. Lots of World Juniors used random pairings.
• Dubov The Dubov system attempts to equalize the strength of opposition of players within the group. It pairs those with the highest rating due Black against those with the lowest average rating of opponents due White. So the effect works for players who have just had White. For those who were Black, their day will come.
• Olympic FIDE has a different pairing system for the World Team Championships, also known as the Chess Olympics. See here for details. In brief, teams are ranked by sum of opponents' scores, then paired TvB.
• Look-ahead Near the end of an event, the leading groups can be sparse, players have already met each other, colours no longer alternating. Sometimes, you can see ahead to the next round that a certain pairing will be required, and change the colours, or even the pairings, in the current round. When you hear about players having to get three of the same colour, or other inequities, sometimes it was because look-ahead was not done, or because the system did not allow look-ahead.
• Suttles. In the 1970s, Grandmaster Duncan Suttles suggested that in many tournaments--especially ones where there is a wide variation in strength of the players and the number of rounds is much greater than 2^n--the leaders meet too early and the logical final-round climax of the tournament is thwarted. He suggested decelerating the middle and perhaps later rounds, but definitely not the last round, so that the top two did not meet until then.
• Phantom. Accelerated systems, including SAD, rely on Phantom Points being given to higher-rated players in early rounds, and later taken away.
• Class. A class is a range of 200 rating points, usually broken at 1400, 1600, 1800, 2000, 2200. In the class method, players rated 1600-1799 (called Class B in North America) would play only amongst themselves. It is used in Class tournaments, and also in the Kerr System.

The Approaches become Systems

As background to the discussion which follows, here are some variations to the standard Swiss pairings.

• Traditional Swiss Pairings use the TvM method. I'm going to call these TSP (nothing to do with a cleaning chemical) to distinguish what is different in the following variations.
• FIDE. The World Chess Federation (FIDE) main system. See details. A flavour of TSP.
• CFC. The Chess Federation of Canada pairing system is TSP and almost identical to the FIDE System. See details.
• Dubov. See details. The 1997 version, which on June 1, 2003 is the same as the FIDE version, is here.
• Class. Each class is separate and has a separate prize fund. This allows players to meet an even field of about their own strength. Drawbacks include that only players in the Premier or Open section can win the tournament per se, and what to do with unrated players, or peer group prizes such as Junior and Female?
• Kerr. Expert and organizer Ray Kerr invented a hybrid of TSP and Class events. Players could declare to play in their own class. Once declared, a player would be eligible for Class prizes but not for top prizes, and would generally play opponents of the same class. Kerr usually made use of Accelerated Pairings, see next item.
• Haley Accelerated Pairings. Canada's Philip G. Haley was (and is) a codifier of Swiss System pairings. To say Haley Pairings would be the same as saying Swiss Pairings. He also developed Haley Accelerated Pairings, which we are going to call Accelerated Pairings. In TSP, the early rounds can be wasted mismatches. It is not uncommon for the rating difference on each board to be 600-800 points in the first round. This in effect wastes a round for the top players, who meet opponents with an infinitesimal chance of winning the tournament. In Accelerated Pairings, the top half of the tournament is assumed to have won a fictitious round 0. So in round 1, players from the top quarter meet opponents from the second quarter. In round 2, losers from the top half meet winners from the bottom half. After round 2, the fictitious round 0 is forgotten. The result is quicker sorting out of the top players and more equitable pairings. Haley introduced Accelerated Pairings into the international arena at the 1970 Canadian Open in St. John's, Newfoundland. I'm calling it HAP.
• Système Suisse Accéléré Degréssif. The French SAD I will translate as Stepped Accelerated Directive. The last word isn't very good, but I have a long list of D-words that are equally bad. This way the acronym is the same in both official languages. Jean-Claude Templeur who has a chess website but does not seem to discuss his pairing system there! directs the tournament at Cappelle-la-Grande. It typically comprises over 600 players in a single-section 9-round Swiss. If you do the math, a standard Swiss is only good for 512 players. It is SAD. En français ici on peut lire les descriptions des systèmes SAD y Haley. In SAD, the field is divided into three groups, given 2, 1, or zero phantom points before pairing the first round. Phantom half-points are added when a player achieves 1.5, 3, 4.5 and 5 points (in a 10-round event), but no player may have more than 2 phantom points. The phantom points are discarded before pairing the last round.

2003 Canadian Open - Kapuskasing

• The 2003 Canadian Open takes place in Kapuskasing, Ontario, as a 10-round one-section event from July 12-20. There is one game per day, except two games on Sunday July 13th.
• This is an open tournament. Anybody may enter and win the tournament.
• There are significant prizes for the top six finishers, plus Class prizes and prizes for the top female players (Women's prizes).

  ACTUAL PRIZE FUND

  Prizes	Total	Open	U-24	20-2199	18-1999	16-1799	14-1599	U-14
  Total	40,000	16,000	4,050	4,550	4,550	4,150	3,450	2,450
  1st	15,500	6,000	2,000	2,000	2,000	1,600	1,250	650
  2nd	9,100	4,000	1,000	1,000	1,000	1,000	650	450
  3rd	4,850	2,000	500	500	500	500	500	350
  4th	3,550	1,500	350	350	350	350	350	300
  5th	2,200	1,000	200	200	200	200	200	200
  6th	500	500
  Women	4,000	1,500		500	500	500	500	500

  The prizes were distributed according to the rules in the CFC Handbook. A piece of advice: make the Women's prizes not the same as any place prize in the same section. Which prize does she take? never became an issue in prize distribution, but it might have.

  By my guess, the total paid entries was about $12,000, so the prize payback was about 330%, thanks to the sponsors, both public and anonymous. A typical not-for-profit tournament without sponsors might have a prize payback of 75%.

• The tournament organizers have many goals, including the possibility of Canadians earning title norms. In the past, Canadian Opens have sometimes advertised that title norms were available, but the use of TSP or even HAP left slim chances for a player having a good tournament to achieve that. In addition, I had my own goal, which was to give players the opporunity to meet opponents of roughly equal strength. Typically under TSP or HAP, a player meets an opponent a couple of hundred points weaker, wins, then plays another opponent a couple of hundred points stronger. The swings, or yo-yo effect, reduce as the tournament ages, but are still significant even in the final round.

  A study of the yo-yo effect in recent Canadian Opens remains to be done, though I have collected the data.

• The Kaps Pairing System has in mind a tournament with 200 to 400 players and upwards of 3 grandmasters. In fact there were 141 players and 15 grandmasters. I made one change to the system because of that.

Generalities of the Kaps Pairing System

Goals of the pairing system, in order of importance:

• any entrant may win the tournament;
• to produce a fair winner;
• to produce fair prize winners and Class prize winners;
• to produce fair winners of the female prizes;
• to provide opportunity to IMs to make GM norms;
• to provide opportunity to players over 2200 to make IM norms;
• to provide the opportunity for those who are doing well against FIDE-rated opposition to achieve a FIDE-ratable performance (based thus on 4 or more games against FIDE-rated opponents).
  The 2002 FIDE Congress in Bled decided that a FIDE-rateable performance for an unrated player should be 3 or more games with at least 1 point. This is a good change, and it was in effect. Unfortunately, FIDE's website did not reflect this important change, we had to rely on insider information to know it.
  56 players had FIDE-rated performances, including 11 unrateds.

The following pairings will be avoided, in order of importance:

• playing the same opponent a second time (absolute);
• for a player who has received a point without playing, receiving a full-point bye (absolute);
• receiving three more of one colour than of the other;
• receiving four of the same colour in five rounds;
• receiving the same colour three times in succession;
• playing a family member; Nobody had to play a family member
• equalizing colours;
• playing an opponent from the same small city; Unfortunately, some D players from the same small places in Northern Ontario had to meet late in the event.
• playing an opponent from the same country (other than Canada or USA) in the last two rounds;
• alternating colours (in an odd-numbered round);
• playing an opponent from the same country (other than Canada or USA);
• playing an opponent from the same large city. This rightly was accorded the very lowest priority.

Although it is not a stated priority, the yo-yo effect will be avoided.

Introduction to Kaps Pairing Techniques

Declaration

A player may declare to be eligible only for Class (including female) prizes. Such a player is then not eligible for top prizes. When the declaration comes into effect, the player will thenceforward meet only opponents in the same class or lower.

• Players rated 2200 or above may not declare.
• Declaration will not affect the pairing in the first few rounds.
• Declaration must be made in writing on the form provided.

  I did in fact accept a small number of verbal declarations.

  In addition, there was a case that I might have handled better. Very early we asked ourselves whether Dina Kagramanov, the only woman rated 2000-2199, would have to Declare in order to win the $500 top woman Expert prize. I thought: she's going to the World Youth, so she should be able to play strong opposition if she earns it. So my answer was "No." Later she asked the question herself, and the answer remained the cryptic "No". As the tournament went, Dina lagged behind the undeclared Experts, and she was paired with Experts or lower for the whole tournament. So, I was surprised when she asked to split the top women prize in the Open section with Pia Cramling, who faced a field over 400 points stronger. As it turned out, Pia's weakest opponent at 2184, was Dina's strongest opponent. Although I sympathized that she had never Declared; I Declared for her.

  Declaration Statistics

  Declaration in Round								--
  Group	Total	5	6	7	8	9	10	not
  Total	101	72	7	5	2	0	3	12
  X 2000-2199	16	7	0	0	1	0	2	6
  A 1800-1999	22	12	3	2	1	0	0	4
  B 1600-1799	21	19	1	1	0	0	0	0
  C 1400-1599	20	16	1	1	0	0	1	1
  D 0-1499	22	18	2	1	0	0	0	1

  A player rated under 2200 could Declare before the 5th or any subsequent round. Declaration made the player ineligible for the top prizes and meant meeting only opponents in the same or a lower class. Undeclared players could still meet higher-class opponents, though they often (eXperts, for example) were paired in their own class anyway.

  Over 70% of the players Declared at the first opportunity. In general, the higher the rating, the more likely a player was to Declare later or not at all.

  Notes: One each A C and D did not Declare and withdrew before the tournament ended. Reasons for withdrawal: called back to work (2), exhaustion (1).

• Once the games for a round are finished, Declaration for the next round will normally close at once, without warning.
• * To be eligible for class prizes, a player must declare before the 10th round. This does not apply to 2200-2399 players, who may not declare in any case.

Norm Pairings

We want to provide norm opportunities for those who have a reasonable chance of success. A little arbitrarily, but also conveniently for the tournament structure, we have chosen 2200 as the lower limit for an IM norm candidate CIM, and 2350 with an IM title as the lower limit for a GM norm candidate CGM. A candidate will meet only other opponents rated 2200+ FIDE, and will get first dibs on opponents with the necessary titles and national affiliations. These considerations will override the Dubov or TvM pairings in rounds 1-8, but not in rounds 9-10 if prize considerations are involved.

First, I did not always give candidates opponents rated 2200+ FIDE.

Second, the idea that I needed to do so was incorrect. See below for detail.

CIM or CGM players may, in writing, choose to opt out of Norm Pairings (so far as their own potential norms are concerned), but they should not assume that this will necessarily give them an easier ride, pairings-wise.

As the tournament develops, candidates might ruin their norm chances (by losing), so we will keep track of the performance ratings, and remove candidates who fall below standard. The standard for an IM norm is 1850 plus 50 points per round played. For a GM norm, it is 150 points higher. Standards will be checked before rounds 3 and following.

FIDE rules state that more than one bye makes a player ineligible for a norm.

Colour Allocation

The normal considerations will apply. Colour allocation may even be important enough to override some of the other considerations here.

Note that while colour equalization (e.g., wbbw instead of wbww) can be important enough to have a player put in a different group, colour alternation (e.g., wbwb instead of wbbw) will not cause pairing panic. A moderate number (say 20%) of pairings that don't alternate can even make later pairings easier.

Women's Pairings

In each class there is a tournament within a tournament for the top women's prize. The winning score is difficult to predict, so it is important that players, even in the middle of the pack, receive equitable pairings. So Dubov is favoured over TvM.

If near the end of the tournament, the top women in a particular class are near the same score (but not likely winners of a class prize), they will be paired against each other.

In the Exact Procedure below, Female (4) means that if four or more leading female players are 1 point or more from the lead in the tournament or in their class, and also within 1 point of each other, they will be paired as a group.

Unlike some of the other elements of the Kaps Pairing system, the Women's Pairings provisions are not likely to affect many pairings.

SAD Groupings

Divisions need to be made at a level which does not make us go through too many hoops to make Norm Pairings. The divisions will be at 2200 and at 1800.

Miscellaneous

• Undeclared Class Players Because of SAD, these players will tend to play higher-rated opponents with lower scores. To preserve Norm Pairings, we may pair them against Class opponents with less points but higher ratings. Finally, if they run out of prize contention, they could be paired against anybody.
• Dubov To provide equitable strength of opposition, the Dubov system will be used in the middle rounds.
• TvM The TSP standard Top-versus-Middle will be used at the beginning and end of the tournament.
• Suttles If the leading group is smallish and the next group is the same size or larger, the two groups will be joined. In the Exact Procedure, below, Suttles (4) means that if the leading group consists of 2 3 or 4 players, and the next group is at least that big, join them. Does not apply to Class groupings.

Here are the pairing methods, round by round:
1. Round 1 : SAD, divisions at 2200 and 1800. TvM.
2. Round 2 : SAD (continued) TvM.
3. Round 3 : SAD (continued) TvM.
4. Round 4 : Declarations recognized. SAD. Dubov

   Declarations were not recognized until round 5. This stemmed from the low turnout (smaller classes would become difficult to pair in later rounds) and the request of a few player. No players wanted Declaration earlier.

5. Round 5 : Declarations. SAD. Dubov. Suttles (4)
6. Round 6 : Declarations. SAD. Dubov. Suttles (4)
7. Round 7 : Declarations. SAD. Dubov. Suttles (3) Women (4)
8. Round 8 : Declarations. SAD. Dubov. Suttles (3) Women (3)
9. Round 9 : Declarations. TvM. Suttles (2) Women (2). Look-ahead.
10. Round 10 : Declarations. TvM. Women (2). Look-ahead.

(Added June 20, 2003). Please refer to: prize page at the tournament official website, which is reproduced below:

Ratings, for pairings and prizes

Details:  For Quebec players who are unrated or inactive CFC, the FQE rating will be used.  Otherwise, (Established) CFC rating will be used for Canadians.  For US players rated below 2200 FIDE, use USCF ratings.  All other foreign players, use adjusted FIDE ratings.  Foreign players without a FIDE rating will have their national ratings converted to FIDE and then 40 points will be added for CFC equivalence.  For other systems, the organizers will add points generously.

Adjusted FIDE ratings are:
Below 2200:  FIDE rating
2200 and up FIDE rating + 40 points

The organizers reserve the right to assign a higher rating to any player whose rating has taken an abnormal drop of 200 or more points.

FIDE Title Regulations (added Jun 20 2003)

FIDE has recently published the new title regulations, effective 1 July 2003 as a zipped Acrobat file.

Here are the norm levels for a 10-round tournament. The figures are the minimum required average of FIDE ratings of the 10 opponents.

Pts	GM	IM	WGM	WIM
8	2381	2231	2181	2031
7.5	2408	2258	2208	2058
7	2452	2302	2252	2102
6.5	2491	2341	2291	2141
6	2529	2379	2329	2179
5.5	2565	2415	2365	2215
5	2601	2451	2401	2251
4.5	2637	2487	2437	2287
4	2673	2523	2473	2323
3.5	2711	2561	2511	2361
Titles	----3 GM----	-2 GM or 3 IM-	2 GM or 3 G/WG/I/F-M	3 WIM or 2 G/WG/I/F-M

FIDE rounds up .5 averages. So sum of opponents' ratings of 26005 is a 5-point GM norm, but 26004 is a 5.5-point norm.

Other requirements about opponents regarding norm eligibility:

• Unrated (entered at 1800) may not exceed 2.

  This part is embarrassing. Unrateds are not entered at 1800.

  It stems from my reading (mis-reading) of the ambiguous FIDE title regulations B 01.

  1.46c For the purposes of norms, the minimum rating (rating floor) for the opponents shall be as follows:

  • Grandmaster 2250
  • International Master 2100
  • Woman Grandmaster 2050
  • Woman International Master 1900

  [...] 1.46d Unrated opponents not covered by 1.46b shall be considered to be rated at the rating floor level.

  I interpreted that the word opponent should fall between Grandmaster and 2250, whereas the correct reading would be the word norm.

  The clarity of the meaning was lost somewhere. In Stewart Reuben's The Chess Organiser's Handbook  2nd edition, the analogous rules are:

  • 9.3 Unrated players are treated as if rated 2000. [...]
  • 9.4 [if the rating system is broadened, which it has been] the lowest rating of each opponent of the norm candidate shall be: GM 2250; IM 2100; WGM 2050; WIM 1900 (GA 2000).

  The GA 2000 version was clearer. They provide a clue, but I read them only later.

  So, my interpretation was that any unrated opponent would count as 1800 (the current rating floor), in other words, an unrated opponent would be a death sentence to IM norm hopes. On the contrary, two FIDE-unrated (the weaker the better) opponents are very good for a norm chance. But in the Canadian Open, I avoided any such pairings like the Plague.

  Eugene Perelshteyn made the point to me during the tournament, but I told him that he had misinterpreted the rule! I am usually open to suggestion, but this was well into a 15-day string of 16-hour workdays. Maybe there should be a rule that the Chief Arbiter can work only 8 hours a day. Hey, I like that! Who writes those rules anyway?

  Here is a shorter and unambiguous wording of 1.46

  1.46 Rating of opponents
  For an example of percentage rounding see 1.44.
  

  1.46a Maximum 22% of the opponents shall be unrated. Initially unrateds shall be counted at the rating floor--which in 2003 is 1800--but see 1.46c.

  1.46b The Rating List in effect at the start of the tournament shall be used.

  1.46c The two (2) lowest rated opponents count at the higher of:
  their rating (for unrateds see 1.46a) or;
  2250 (GM norm), 2100 (IM norm), 2050 (WGM norm), 1900 (WIM norm).

  However, I don't think that 1.46 will last too long in its present format, even with the ambiguity removed. There will be too many 10-player norm tournaments comprising 3 GMs, 5 legitimate candidates, and two players rated 18xx.

• At least 5 must have the (W)GM, (W)IM, or (W)FM title.
• At least 4 must be from other federations than the player's.
• Foreigners may not play more than 7 Canadians.
• There may be a requirement (1.43) for two federations other than the player's, but we will probably nullify that by exceeding 1.43e.

Among the changes in the new rules:

• the player's own rating no longer counts: every norm is a performance norm
• 27 games required, not 24.

URL: This web page is:
http://members.shaw.ca/berry5868/kapn.htm
Last modified February 9, 2005