Reflection time and the length of tournaments
by Wim van Vugt
(posted 31 May 2007)
Wim van Vugt (photo by Frank van der Wolf)

Recently a lively discussion has arisen again
about the best time schedule that can be applied for ICCF
email and server games. In my view there need not be any difference
between these two forms of correspondence chess. A characteristic
of CC is that the reflection time is counted in days, not in
hours, minutes and seconds. The current 8pm
rule (see 6c) could easily be changed into a server analogue,
saying that if a move is sent back within 24 hours ZERO days
are counted, and so on. Such a unification of the rules would
be my preference. But even if this slight rule difference remains
between the two types, one can still look at how the allotted
time is spent and saved up.
A few months ago I was asked if I had some ideas
what length of tournament would be achievable under the current
ICCF
rules of play, i.e. 50 days per 10 moves. During this time
another item was discussed thoroughly on TCCMB:
the socalled DMD (dead man's defence). DMD can be described
as an extreme slowing down of the game in a completely lost
position. All the accumulated time is used to postpone the inevitable.
Under the current rules all unused reflection time is carried
over to the next group of 10 moves. It is not unusual for 200
days to be accumulated by move 40.
There are two different viewpoints if one asks
"How long does a tournament take?" Firstly there is the viewpoint
from the position of the organiser. He wants a clear and easy
to handle end date, say two years after the start. Ideally there
then are no unfinished games remaining that need to be adjudicated.
However, a too short end date will result in many unfinished
games and a late end date makes players wait too long for decisions
like promotion and the start of the next round.
The other viewpoint is from the position of the players. They need a reasonable time allotment and most of them prefer a steady and continuous flow of moves. Although it's everybody's right to use one's time as one likes, it's generally felt as bad behaviour and a show of bad sportsmanship if a game is suddenly delayed by many weeks per move as soon as a lost position has arisen.
But does the DMD influence the general length
of a tournament? If not then organisers don't feel much drive
to ask for a change of the rules. If it does then a counter
measure is to be expected. Several proposals have been made
to stop players from abusing a huge accumulation of time, employing
it to delay the game needlessly. One idea recently proposed
is doubling the time used after 15 days, a method formerly used
in postal chess and currently still in force within the IECG
organisation. Another idea often repeated on both TCCMB
and the new ICCF
forum is limiting carryover until a maximum of 100 days
is accumulated. Another idea well worth considering is connected
with the use of the Fischer clock: 30 days at the start plus
3 days per move, with maximum of 60 days reflection time.
From the viewpoint of organisers another quite
revolutionary idea has been proposed: give 300 days in total
for the whole game to each player and let the players decide
for themselves how this time is distributed over the course
of the game. This looks simple and it solves the organiser's
problem completely, but it creates another set of huge problems.
We could see equal games continued on and on as one side attempts
to win on time, a clear conflict of FIDE rules (art. 10.2).
See the FIDE
Laws of Chess section E101A and go to paragraph 10.2.
Players who want to stall their game(s) when prospects become worse have often played quickly in the beginning and are left with 200250 days accumulated time that can be used to delay the end for almost a full year. Any savedup time more than 100 days will be abused sooner or later so that the "300 days per game" idea will miss its goal.
It would be a serious mistake to experiment with this idea in a successful tournament like the Champions League. It could ruin that tournament in one season. In short, the "300 days per game" may be an easy solution for organisers but would create abuse and new problems for the players and should not be considered.
In order to answer the question of how long
tournaments should take I decided to do a more fundamental research
about the length of games, expressed in number of moves and
time consumption per move. In order to get a reliable statistical
distribution I have used some 48,000 ICCF
games. When analysed with ChessBase it's easy to get an output
showing the whole distribution. The result turned out to be
a slightly asymmetric skew distribution with the following parameters:
mean number of moves: 37; standard deviation: 15. And, most
importantly, 95% of all CC games are shorter than 60 moves.
Figure 1
You would expect to base your reasoning on this figure since 60 moves takes 300 days per player and 600 days for both, at maximum. Leave must also be added for both players (30 days per year) so 720 days are needed. For a twoyear tournament no time would remain for scheduling the next round, to perform adjudications for the approximate 5% of remaining games and for the players to prepare themselves for the next round. Therefore, an extra three or four months might be needed causing us to conclude that only a 2.5year cycle can be reached.
But, before going into more detail about the characteristics of the time distribution curve, let's first look at the 95% criterion.
This means that 5% of the games are not finished and must be adjudicated. In an 8 man tournament this means 12 games, but in a big 13 man qualification group the number of games to be adjudicated becomes 4. This is a too high figure, in my view. At most one game per qualification group (78 games) should be acceptable. This boils down to 1%. Or 99% of the games must be finished by the end date.
When looking at the more rigid requirement of 1% it turns out that the time distribution is remarkably less symmetrical than the game length distribution. What's important is not that we know in how many moves 99% of the games are ended but in how much time 99% of the games are finished. Here also must be taken into account that in longer games it sometimes occurs that more than 6 days per move are used as an average, due to the accumulated time. A general effect is that in the beginning about 3 days per move are used as an average while at move 60 this has become about 6 days or more.
Figure 2
Making use of data from about 1000 games, most
of them coming via TCCMB
from players who have sent me their data by private communication,
I have been able to combine both variables (time and length)
into a general time distribution curve for correspondence chess
games.
Figure 3
From this time distribution well founded conclusions can be made.
Some characteristic values are:
Mean = 250 days; standard deviation = 170 days.
But more important are the following values:
95%

96%

97%

98%

99%

593 days

630 days

675 days

730 days

800 days

It is easy to see that the time distribution is significantly more skewed. The centre of gravity is shifted to the left, caused by a combination of short games and little time consumption per move. The average time consumption per move is not constant. The more moves a game takes the more time is invested per move. In the beginning of the game about 3 days per move are used. In this stage a large accumulation of time is generally built up. As the game progresses, the time consumption increases supported by time accumulation.
If no more than 1% adjudications is desired, a tournament must take 800 days. This is 2.2 years. But if these 800 days are spread over three different calendar years, then 3x30x2=180 days must be added for player leaves, making a total of 980 days. This is 2.7 year. The remaining 4 months can be used for adjudications and planning the next round so that a 3year cycle is the best one can achieve under the current rules.
It is suggested that the increasing time consumption keeps pace with the complexity of the position. This sometimes is true. But it also happens quite often that there is no relation at all between these variables. A complex position in another game or simply an overload of many games may cause a player to decide to stall a game or several games only to provide some breathing room to spend extra time on one particular game.
Yet it is not a healthy system when such enormous
amounts of time can be accumulated as is now the case within
ICCF. A buffer of some 100
days must be sufficient to cover the whole spread in time consumption
as displayed in figure 2. Larger time consumptions have been
reported several times on TCCMB
but always in connection with completely lost positions. From
my own experience I can confirm that it is a very frustrating
phenomenon. Not everybody considers it an abuse of the playing
rules when moves are sent once a month in a lost position. But,
in my view, it's not just making use of one's right to manage
one's time as one likes. It is bad behaviour and it should be
made impossible by the rules.
Looking at the time distribution (figure 3)
one can conclude that the tail at the righthand side is caused
by a combination of long games (many moves) and a high time
consumption per move. The former cannot be prevented but the
latter certainly can. There are a few simple measures that do
not harm the normal player but prevents excesses on the high
time consumption side. It isn't necessary to speed up all games
if one likes to speed up tournament cycles. One only needs to
speed up those games that cause the delay of the cycle. Therefore
a change from 60/10 to 40/10 will not result in a change from
a 3year to a 2year cycle, because it's the tail that causes
the delay.
The first necessary measure is to limit carryover time to a maximum accumulated reflection time of 100 days. A healthy buildup of a reserve is still possible and it does not harm normal play, but it makes an extreme DMD behaviour impossible.
A second necessary measure is to limit the total leave that can be taken. Under the current rules each player may claim 30 days leave per calendar year. This means that if a tournament starts in December one can take a free leave for the whole month without any consequences. The same applies to a lesser extent when the tournament starts in the second half of the year. An improvement to shorten the total duration of a tournament could be that the "30days per annum" claim is maintained but combined with a tournament maximum of 60 days. This would lead to a tournament duration of 800+120=920 days (2.5 years). The remaining half year can be used for adjudication and planning the next round. Under these circumstances there also may be reserved some time between the announcement of the tournament to the players and the actual start date. A preparation time of two months seems desirable for the players, certainly when high level competition is at stake.
An additional measure could be to look for a
more strict formulation of rule 3b. The previous wording (i.e.
the TD must agree to a longer delay) was better than the current
one, in my view. And a change from 40 days to 30 days in that
article 3b would be welcome. It would make a simple ICCF
standard: 30 days as a maximum for leave per annum and a maximum
for days to be allowed to spend on one move.
Finally, a last idea to speed up games while
a "normal" tempo can still be maintained is the following Fischer
clock idea: 30 days at the start plus 3 days per move, with
a maximum of 60 days reflection time. Then the previous three
proposed measures can probably be omitted. In any case it's
a good idea to experiment with it sometime.
© 2007 Wim van Vugt. All rights reserved.
