Forum for Timing Systems - Redux
modifying fischer time? [#820]
On the page it is suggested that you could either introduce a maximum thinking time per move or cap the time accumulation.
My opinion on this is that if you want to do this, you have been using the wrong settings to begin with, or Fischer time is not what you need. I believe that Bronstein can cover up those cases.
A cap for the time accumulation is already used, for example, at the DGS, where your accumulated time can never exceed your initial time. A typical setting is 30 days + 1 day/move, never exceeding 30 days.
I think that this is the wrong approach. You should calculate how long you want the game to be, then use an appropriate setting. In the DGS example, you might want a game to last at most 180 days. Possible settings would be 40 days + 8 h/move, 27.5 days + 10 h/move, or 15 days + 12 h/move.
Of course, you might have the concern that on such a server one player might login only once per day while the other makes his move always a few minutes afterwards. In such a scenario, the players are spilling the time themselves already, and Bronstein or the like might be a more appropriate setting to keep it fair.
IMO real-time play, for example in tournaments, and turn-based play such as DGS should be treated separately. A timing system for real-time play controls the actual thinking time, and for example in tournaments, it is important to be able to set a maximum time a game can take. For turn-based play a timing system adjusts how often you are supposed to log on the server and make a move. In friendly turn-based play the main purpose of time control is not to end a game within given time, but to ensure that the game progresses at a reasonable pace (so that the players do not get annoyed because the game stalls).
In turn-based play, I prefer Dragon-style capped Fischer 10d+1d. Both parameters, then main time and the added time per move have natural interpretations in this context. (1) The main time is the time you are prepared to wait for a game to time out in case your opponent abandons the game and will not move anymore. (2) The added time is the time you're prepared to give your opponent, in average, to answer your moves.
agreed. Two aspect of timed games allow for clock skill to provide one player with advantage:
- time crunches (most apparent in absolute time but also common in Canadian)
- time spilling
The whole point of Fischer Timing is to reduce these opportunities for employing clock skill as much as possible. Adding any sort of time spilling to Fischer Timing is not an improvement.
The funny thing is that the main hurdle to accepting Fischer Timing appears to be lack of familiarity. What makes this amusing is that I often spend a fair amount of time explaining Japanese and Canadian byo-yomi when I teach newbies on KGS.
Fischer timing is actually much easier to explain than Japanese byo-yomi.
I must conclude with saying that I'm happy that my biggest complaint about KGS is something as insignificant as not implementing Fischer Timing. Given that Japanese and Canadian byo-yomi are available this is really a minor complaint!
Perhaps I expressed myself poorly so far. Let's assume that there is sufficient difference between online Go and over the board Go to warrant a new purpose for online in addition to the two set out in the Background section.
Purpose #3: Ensure a lively pace to the game.
Here by lively pace I mean that plays happen at reasonable intervals with no long gaps in between. The standard way to address this is to simply play blitz (= short time limits under any time system). However, blitz and "lively pace" as I have defined it here are two separate ideas. The question that I wanted to raise with an MTT or Fischer cap is whether we can achieve lively pace without resorting to blitz, given that we have complete control over how the clock works? In other words can we achieve lively pace while avoiding time scrambles, etc.?
Harleqin has recommended Bronstein timing instead. Let's look at what this might mean. Below are several graphs that I prepared using the timings in Game 2 presented on the main page. That was a 60-minute main time game where Black and White each used about 48 minutes (I picked Game 2 because there were no really long-thinking plays by either player and the long short plays were distributed throughout the game). In the first graph I have simulated the remaining time for Black throughout the game assuming:
- Bronstein timing with initial time of 20 minutes (1,200 seconds) and 30 seconds increment,
- Fischer timing with initial time of 5 minutes (300 seconds) capped at 300 seconds and 30 seconds increment,
- Fischer timing with initial time of 5 minutes (300 seconds) and 30 seconds increment.
Now what should we take away from this graph?
- First and foremost, I think we should understand the implications of the Bronstein line. At these settings, Black would have run through the initial (unconditional) time around the 215th play in the game. Thereafter Black would be playing on 30 seconds per play until the end of the game. This is simply the remaining weakness in Bronstein, the non-spilling initial time helps but the spilling increment means that the time remaining will always be a one-way street, heading down to zero. At that point it reverts to single-period byo-yomi. Personally I do not see Bronstein as a final solution to any problems encountered in using a clock to play games. Note that the difference between the Bronstein line and the pure Fischer line represents the amount of spilling that occurs in Bronstein with this set of timings per play.
- Second, the Fischer capped and pure Fischer lines illustrate the effect of the cap when the Fischer increment turns out to be larger than the average (mean) time per play in the game. In this case I knew ahead of time that the real average was about 22 seconds per play (see the histogram graph on the main page). In such a case, pure Fischer remaining time builds up throughout the game. The cap prevents this as shown in the graph. The problem is what does this mean? The answer is of course - Nobody knows! We can debate it as long as we like but we can not arrive at any conclusion without being able to put it to the test. :-)
The second graph is based on tighter time limits - the increment is reduced to 20 seconds for all three systems. As mentioned above, the normal way of striving for pace is to shorten the time limits. When we do that here, we see that we run out of time (reduce to byo-yomi) that much sooner under Bronstein and the cap on Fischer becomes meaningless as the Fischer increment is insufficient to build the remaining time over 5 minutes until the very end of the game.
I guess the real question between Fischer and the cap comes when we compare different timings that might represent alternative ways to try to achieve lively pace. The third graph shows Fischer with a 20 second increment compared to capped Fischer with a 30 second increment.
The point is whether the benefit of being able to use a larger increment without having the accumulation of time get away from the players (if that indeed is a threat to the pace of the game) is sufficient to justify a modification to the basic system. A careful examination of the graph will reveal that the cap's higher increment allows a faster recovery of available time after it is run down by thinking on a particular play. Again we can have opinions but they are actually uninformed in the absense of experience.
That faster recovery is the counterpart to the faster dropping to zero, both of which makes sense under the purpose of avoiding long thinking times. However, I would not regard that additional purpose anywhere near as necessary as the core two, regardless of online or over-the-board play. If any at all, I rather can see at least three supplemental objectives we could likewise consider:
- (#3a): structural simplicity / low entropy (avoiding the need to check and possibly adjust to the clock too frequently)
- (#3b): steady pace (avoiding long thinking times for single moves which the opponent might feel uneasy about)
- (#3c): protected free minimum time per move (for fairness towards physically languid players and/or slow connections, which suggests another way to mix Fischer with Bronstein)
I think, too, "Fischer cap timing" might really be one of the most sensible hybrids: with the cap set at the initial time, the system works like Fischer as long as you're below, and like Bronstein when you're above what you started with.
Nevertheless, it lacks the pure systems' elegant simplicity. As all enhancements over Fischer or Bronstein I have seen suggested so far, this conflicts with #3a: In order to suitably switch between Bronstein-optimal and Fischer-optimal time behaviour, the players have to check the main time and compare it with the cap frequently. The time game gets more complicated thereby. Of course, the main time has to be checked against zero from time to time anyway, but the cap introduces a second criterion to evaluate and adjust to. This requires more attention than either pure system, since proper time usage under "Fischer cap" cannot be internalised like a simple Bronstein delay or the even simpler "the faster you move the bigger your savings" Fischer principle can.
protected free minimum time per move (for fairness towards physically languid players and/or slow connections, which suggests another way to mix Fischer with Bronstein)
I say "to hell with protecting the physically languid". They simply should play with longer time limits. Short slow uncoordinated people have no business trying to play professional basketball and languid people (or people with slow connections) have no bussiness playing blitz games.
BTW, I happen to be one of the languid people (e.g. I prefer 50 second or longer byo-yomi periods). I don't need a Bronstein subsidy for the slow, I simply chose longer time limits.
I believe KGS has the client measure elapsed time so your clock doesn't start ticking until your opponents move has arrived (can anyone verify this)?
Of course DGS is the ultimate solution for people who are slow (for what ever reason).
![[Welcome to Sensei's Library!]](images/stone-hello.png "Sensei's Library")