![[Welcome to Sensei's Library!]](images/stone-hello.png "Sensei's Library")
Thank you very much for your work, Dave, especially for your clear concepts and for the informative graphs.
Combining time systems with an "absolute main time" has an additional drawback: it is not absolute, because the player doesn't lose at the end but only enters the next phase. Instead, I'd like to call it "free time" because you don't need to move at all during this time.
This leads to the known phenomenon (like you said, human nature...) that some players only play about 20 to 30 moves during their "main time" and the rest of the game in "byoyomi". This means that the game can last significantly longer than the tournament organizers intended, causing round delays.
To reduce this undesired effect, some tournament directors have increased the byoyomi speed significantly ("I'll teach you to use your main time"). The result is that the time scramble at the end is not avoided, just formalized - at some point, even absolute timing would be better. Obviously, they should rather reduce the free "main time" which causes the problem.
I also agree that having distinct phases during the game is a bit unnatural. Such time systems have an undesirable impact on the game - sometimes you can actually see the point when one player entered byoyomi in the game record. Having just one big phase with the smooth delivery of time like in Fischer is so nice in contrast.
I'd like to point out that "spilling" actually is undesirable, like you concluded. It should be a principle that every second that you do not use for thinking now is preserved for later. Otherwise, players are interested in withholding their plays to avoid this spilling. This can not be in the interest of the tournament director. Also, players should be relieved as much as possible from time management so they can concentrate on the game.
Thanks Harleqin, complements are always appreciated :-)
Have you acted as tournament director or observed the problems as a player? It would be interesting to see a little more on the problems the TD's are trying to deal with. How much time is typically scheduled between rounds today, using what time limits? What estimates for total game time are assumed in deciding the schedule and how frequently do delays develop?
Have you participated in any tournaments run under Fischer timing? If so, what kind of limits were used?
I have not played in a Fischer timed tournament yet, but I hope to be able to attend to one in the near future. I'll be happy to report on others' as well as my experiences.
For now, I have only a few club games I played with Fischer time to test it, and a few ab initio thoughts about what would be sensible settings.
Firstly, I'd like to point out the very simple calculation of the maximum duration of a game. I assume the whole game to be 300 moves long, which I estimate to cover over 90% of all games. The total time delivered to the players after 300 moves I call t300. As an example, for 20/20 (20 min + 20 sec/move) t300 = 140 min or 2:20 h. This could be a good setting for a tournament with the rounds' starting times set 2:45 h apart (25 min left for preparing the next round, making wall lists, getting the players to their boards etc.). To be able to do this calculation is very important for tournament directors.
In my slow club games with Fischer timing, the main advantage seemed to be that you have to watch only one clock over the whole game, without having the pressure from an absolute end, and without entering different phases. Fischer is not about building or using "reservoirs", you just have a very understanding and nice clock ticking while you can concentrate on playing.
I also played some blitz games with 1 min + 5 sec/move, or sometimes even 4 sec/move. Normally I don't like blitz games because they are usually played with absolute time, e.g. 10 min. I don't feel comfortable then since I don't know how much time I can use for a move at all. With the above Fischer setting, I can just play, keeping a little eye on the clock but never having to calculate the whole remaining game time, so it is really comfortable - apart from being fast, of course ;)
As for sensible settings, I think a certain relation between initial time and bonus time should be observed. Bonus time sets the pace of the game, while initial time adds flexibility. I guess settings like 1/4, 1/5, 2/6 make good blitz settings, 10/10 or 12/12 might be fast settings, 20/15, 20/20, 30/20, 30/30 could be used for normal tournaments, 40/40, 50/40 and more for national championships and the like (always initial time in min/bonus time in sec). The big japanese title matches could perhaps use 6 h + 1 min/move (t300 = 16 h) ;o).
Dave, this is great work! :-)
I think that your idea of studying games where neither player entered overtime is sound. How do players naturally use their time? As you indicate, in this era of digital, programmable timing, we can tailor the timing system to the players rather than to the clocks.
True, there may be a bias towards slow players, but it is reasonable to think that a system that fit them would be acceptable to quicker players with shorter time limits. In any event, with games where players enter overtime, the type of timing system may affect how they play. Then it is hard to guess how they would have played under a different system.
@dave: Bravo for the excellent work! While several people have come to the same conclusion via gedanken experiment, your empirical data adds more weight to the argument.
A while back I had suggested on the SGF page that a property be added to document elapsed time. There does not seem to be much interest in adding this to SGF or, more importantly, to any of the Go Servers. Too bad. If such a property was available you (or others) could collect meaningful data about byo-yomi time usage.
@wms: I'm not sure what you mean by bias towards slower players.
Spilling time systems are biased against those who, on average, spill lots of time and biased in favor of those who learn to milk the clock and minimize spilling or who play so fast the game is decided before they ever enter byo-yomi.
If a player plays so fast the game is decided before she ever enters overtime then the choice of time system has no real impact on her. Of course the parameters of any time system can be ratched down until they impact even the fast players (e.g. the folks who play 10 second byo-yomi games).
It is possible for both players in a game to be slow i.e. enter byo-yomi early in the game but for one player to be adept and milking the clock while the other is not.
Now I know it wasn't Clinton who wrote that post ... ;)
@jonathan
I looked on the SGF page and the home page at red-bean. I could not find clear indication of what should go into the time left (BL, WL) properties. For research purposes it would be enough if these held the real remaining time before increments. In other words the time left after, for example, the 99th stone would not equal the time available for the 101st stone if Black were in byo-yomi. Today the property holds the time allowance for the following play, at least in KGS files. Does anyone know whether the same holds true for IGS files?
Wow, well put. I can't help but hope this page will get the consideration it deserves.
*agreed*
I never did understand why internet go servers weren't more Fischer time friendly. If I was implementing a timed two player game, Fischer time would be one of the first I'd think of / implement.
Perhaps the the online Go world is so needlessly conservative because it widely sticks with the pros which in turn stick to (offline) tradition? I bet as soon as a major pro tournament introduced Fischer clocks, things would change in the blink of an eye.
Beside the objective effects of time constraints discussed already, they also differ in their demand for the players' attention. Proper time management requires taking account of the clock situation under all time systems, but some require significantly more adaptation than others. Canadian overtime, for instance, performs particularly bad at this, since the remaining time per move is subject to drastical changes throughout the game. Amongst the established Go time systems, I prefer either japanese byoyomi with a large number of periods or absolute time. Fischer would be smoother than both.