Commentary on the Korean 2013 Rules

[RobertJasiek]

asura, for your Japanese rules model, start testing the second part of the official commentary, except examples 16-18. All the other examples must work as shown. Besides, test snapback at a territory boundary and further fake dead stones on the inside / outside.

Do you want to replace local-2 by something else? Why? Recall the equivalence relation to two-eye-formation:
http://home.snafu.de/jasiek/wagcmod.html

Japanese / Korean locality: the complexity is not the same, but the complexity class is the same: first define "force", then define life and death and locality concepts by using force. (My Simplified Japanese Rules and Simplified Korean Rules forgo such and rely on much simpler, implicit localities.)

[jts]

Brief comment:

"The number of liberties of a contiguous group is the sum of liberties of each stone in the group" (p. 5).

That statement is false, and they even give an example without noticing that.

I assume they mean "the number is the sum of (liberties of each stone in the group)", rather than "the number is (the sum of liberties of) each stone in the group". Since a liberty for them is an intersection (a spot on the board, not a characteristic of a group), it makes sense that they are thinking of a collection of intersections, rather than of sigma(liberties of stone i)_i=1ⁿ

Of course they were thinking about a collection of intersections, but that's not what they said. "(the sum of liberties of)" does not parse. Besides, summation is not the process by which we find the number of liberties of the group. Counting is.

What they meant, I suppose, is that the set of liberties of the group is the union of the sets of liberties of each stone. The number of liberties of the group is the cardinality of that set. OC, by then they have lost their audience.

I will grant you that, but I think you are asking for a lot from a document that was clearly not produced by a native English speaker!

[RobertJasiek]

Version 2 of my commentary contained a typo in the Simplified Korean Rules, Game End Procedure, point 2, where of course "two successive," must be "two successive passes,". This is corrected now in version 3.

[John Fairbairn]

"The number of liberties of a contiguous group is the sum of liberties of each stone in the group" (p. 5).

That statement is false, and they even give an example without noticing that.

It is of course possible that it was a straightforward booboo, but experience tells me another explanation has to be considered as potentially more likely. All the Oriental go languages have the characteristic that they do not normally distinguish singular and plural. That works, but sometimes they need to be more specific. There are several ways but one is to use 各, which is sometimes used to mean 'each' but which is also often used simply as a pluraliser. Further, the mathematical sounding 'sum' is probably only the word for 'total number'. So the original sentence could (or 'should', with high probability) have been translatable as something like 'the number of liberties of contiguous stones is the total number of liberties of the stones as a group.'

[asura]

asura, for your Japanese rules model, start testing the second part of the official commentary, except examples 16-18. All the other examples must work as shown. Besides, test snapback at a territory boundary and further fake dead stones on the inside / outside.

I said I tested it on a (relative) small number, but of corse I tested it on the examples of the official commentary, else I had said I tested it not at all

BTW my model works for example 16-18, too, as my motivation was to get a model that ALWAYS works correct. (Of corse I mean the *intended* result, not the flaw with endless ko-passes.)
Of corse (as far I'm aware) it covers snapback and dead stones on the inside, too. What exactly do you mean with "fake" dead stones?
And even a unfilled ko (capturable-3 in j2003) works without introducing a more complex method.

Do you want to replace local-2 by something else? Why? Recall the equivalence relation to two-eye-formation:
http://home.snafu.de/jasiek/wagcmod.html

I'm neither using different local-n's nor using 2-eye-formation. Only can (force) capture or cannot (force) capture matters. Depending on this every string gets exactly one of two states: dead or alive. This is more easy than in j2003. The most difficult is to restrict the area where a new played compensation stone affects the state of the string in question.
This point makes a different with this rules.

Japanese / Korean locality: the complexity is not the same, but the complexity class is the same: first define "force", then define life and death and locality concepts by using force. (My Simplified Japanese Rules and Simplified Korean Rules forgo such and rely on much simpler, implicit localities.)

I dont't see why you think that your model can prove that the complexity class is the same. That would require to prove that your model is the one with the lowest complexity for both rules.
Lets say my model for j1989 is less complex than j2003 but it doesn't work for the "Korean" rules then the reverse would be proven.

edit: Actually this wouldn't prove the reverse, because for this I would need to prove my model is the one with lowest complextity for j1989.

[RobertJasiek]

If your model is so good, where are your rules?

There can be strong doubts whether the writers of the official commentary really understood their examples 16-18. They can have more sequences than are apparent at a glance.

Fake dead stones behind a snapback are stones whose status is unexpected due to the presence of the snapback.

What are your definitions for "can (force) capture", "cannot (force) capture", "alive", "dead"?

Complexity: maybe. There is too little research about it.

[asura]

The rules are in my head and parts of it divided on different papers. If you're interested I could (mainly a matter of motivation) write them down - I need to search the papers and also check which versions are the correct ones as I changed a lot of things during the creation (and it's over 1 year ago). Or maybe it would be more efficient to write them new...
I would be interested to get some feedback, too, because I'd like to know wether they are really good or a failure. (Sometimes when I make a thinking mistake and miss the right point to notice it, it becomes very hard to notice it later on because I repeat the same mistake without thinking about.)


For the examples 16-18 it's quite obviously that they overlooked the cycles containing ko-passes, but (for me) it's very obviously that these cycles are not allowed. Also there are some cycles without ko-passes that can also cause simular problems and need to be solved, too. My method here covers both in the same way.

I don't (explicitly) define "force" or "hypothetical-sequence" as you did in j2003.
Instead my rules uses exactly one playout for each string. Every playout ends in finite moves. Depending on the end position of the playout the string gets the state of either dead or alive. (For cycles it's a bit more complicated because then the moves played are also relevant).
Also the result of a playout doesn't affect the result of other playouts, so the order is not relevant.

OTOH in j2003 the result of capturable-1 also effects captuable-2, what is more complex.

Another advantage with my playout concept is that I designed it to give a "natural strategy". This means it's never bad to kill what you can / live where you can. In j2003 you was forced to define "hypothetical sequence" because there are cases where if you would make a playout it could be good to let capture stones that are capturable-1 or uncapturable (or not capture stones that are dead).

[Bill Spight]

"The number of liberties of a contiguous group is the sum of liberties of each stone in the group" (p. 5).

That statement is false, and they even give an example without noticing that.

It is of course possible that it was a straightforward booboo, but experience tells me another explanation has to be considered as potentially more likely. All the Oriental go languages have the characteristic that they do not normally distinguish singular and plural. That works, but sometimes they need to be more specific. There are several ways but one is to use 各, which is sometimes used to mean 'each' but which is also often used simply as a pluraliser. Further, the mathematical sounding 'sum' is probably only the word for 'total number'. So the original sentence could (or 'should', with high probability) have been translatable as something like 'the number of liberties of contiguous stones is the total number of liberties of the stones as a group.'

IMX, such a booboo is more likely. (And I try hard to come up with interpretations that make sense. I considered that "sum" might be better translated as "total", for instance.)

Unfortunately, we do not have the original Korean, but here is the relevant context.

An empty intersection immediately connected to a single stone or to a contiguous group of stones is called a liberty. . . .

A contiguous group of stones is a group in which each stone is immediately connected to other stone(s) with the same color. The number of liberties of a contiguous group is the sum of liberties of each stone in the group.

Your suggestion, 'the number of liberties of contiguous stones is the total number of liberties of the stones as a group.', is tautologous, given that they have defined contiguous group in the preceding sentence. Not that people do not state tautologies, but to do so here would violate the pragmatic rule against telling us what we obviously know. It would not only be, as Robert points out, superfluous, but a waste of the reader's time.

Leave out "number" and change "sum" to "total" (or maybe "totality") and we get "The liberties of a contiguous group are the total of the liberties of each stone in the group." That is something that we could deduce, but it is not just a tautology. I think that "total" allows us not to count the same liberty twice. The problem arises when we talk about the number of liberties. That is one kind of mistake that non-mathematicians make, assuming that numbers add up when they don't. I am afraid that "number" is in the original Korean.

[RobertJasiek]

asura, of course, I am interested in your rules.

Why do you call it obvious that cycles including ko-passes should not be allowed?

Using one finite playout (per string) is much nicer for application, of course. However, it contradicts the spirit of official Japanese rules, which seek perfection in omniscient play. OTOH, if the players make correct hypothetical choices during the one analysis sequence per string (and possibly per status), then "force" is not needed. You can then (roughly) define that a string is alive if a same-coloured permanent-stone / at least one same-coloured stone is on one of its intersections at the end of the analysis sequence for the string. Straightforward; this should work. I would have to think a bit why / whether local-2 could become superfluous in this manner. But anyway, somehow in this manner one gets pretty / possibly extremely well behaving Japanese model rules. Only a model, because THE PLAYERS MAY MAKE HYPOTHETICAL-STRATEGIC MISTAKES during every analysis sequence! (If you want to prohibit such mistakes, then you need "force" again.)

For cycles, simply end an analysis sequence on a move recreating a position / an earlier position during this analysis sequence.

I do not understand yet what you mean by "natural strategy".

[asura]

I call it's obviously because the given result in the commentary. Also there can be cycles with ko-passes where without ko-passes no cycle would be.
If this is not intended, why did you change the ko-pass rule in j2003 to get this results?

I know that a playout is not exactly the same spirit as j1989, but I wouldn't call it contradicting it. Actually I could simply add your definitions about "hypo-sequence" and "force". And applying the playout-rule can be done just by talking. However this is only a theoretical difference, because under j2003 in practise you only look at some variations and can make mistakes, too, no matter what the rules say (Actually one could say you "hide" this fact with your rules.)

(roughly) defining that a string is alive if a same-coloured permanent-stone / at least one same-coloured stone is on one of its intersections at the end of the analysis sequence for the string is NOT enough for the strings that are capturable-2 under j2003. That's where my "local" definitions comes in play.

In my model I use two different types of cycles : fighting-cycle and disturbing-cycle (well and simply defined what is what so it's a bit more complicated, but that is no problem and only matters in rare cases. As a benefit it's not needed to lift all ko-bans with a ko-pass.


With "natural strategy" I mean that for ALL strings it is always better or equal for the attacker to kill the string and always better or equal for the defender to live with the string.
This may sound trivial, but it is not.


edit: Now after saying how good (hopefully) my rules are it seems I have no choice but need to write them down because else it would look more stupid than a total failure would look
The week arround Christmas I have no PC and no guitar, so I plan to do it then. Motivation and lazyness might change it though...

[RobertJasiek]

Allowing cycles with ko-passes, the intended outcomes are confirmed. (IIRC, cycles with ko-passes are not a problem for examples 16-18.)

The J1989 ko-pass rule needs to be changed, because it is incomplete. See the flaws

022 o 7.2 ambiguous scope of application of the ko-pass rule
023 o 7.2 ambiguous difference between pass and ko-pass
024 o 7.2 unspecified successions of pass / ko-pass
025 o 7.2 ambiguous consequences of alternating sequence on ko-passes
026 o 7.2 unspecified presupposition for early ko-pass
027 o 7.2 unspecified validity of double ko-pass
028 o 7.2 ambiguous "temporarily disappearing" ko

http://home.snafu.de/jasiek/j2003com.ht ... e1989Rules

More importantly, I changed the J1989 ko-pass rule, because its application contradicts the intention of J1989! I forgot the relevant examples, but presumably you find them here:

http://home.snafu.de/jasiek/j1989c.html

J2003 is not designed for incomplete application in practice, but for complete application in theory, even if this should require more storage than the universe offers;)

Can you please show examples
- of capturable-2 to demonstrate why your rules still need some locality concept,
- fighting cycle according to your definition,
- disturbing cycle according to your definition,
- how not all bans need to be lifted?

(Around Xmas, if you like.)

[asura]

With "cycle with ko-passes" I mean a cycle that contains one ore more ko-passes. But it's NOT required to contain only ko-passes.

As far as I can see this is the main problem with the ko rules in j1989 and the (one and only ? ) reason to introduce the ko-pass for all kos from both players.

It's some time ago that I read the commentary to j2003 and IIRC everything that you wrote seems to indicate this, too. (I remember it because I was not happy with the ko-pass for all kos instead for a specific ko from the very first time and payed attention to this)

With a strict application of j1989 rule all the examples 16-18 wouldn't work.

Can you please show examples
- of capturable-2 to demonstrate why your rules still need some locality concept,
- fighting cycle according to your definition,
- disturbing cycle according to your definition,
- how not all bans need to be lifted?

- there are different classes of what is capturable-2 in j2003. For some classes I could avoid "local". I need it e.g. in Life-and-Death Example 2 from the official commentary.

- fighting / disturbing cycle is a bit of a word play related to ING rules:)
(a little bit related) but here it simply means: a cycle is a fighting cycle if it contains one ore more intersection of the string thats state is in question,

- else it is a disturbing cycle.
(When the game is already over the ING rules works much more easy

- as I said above the problem with the ko-pass in j1989 is when they make endless cycles possible. The most flaws you have listed are that they are not very clear formulated. But while this can be easyly fixed the cycles are a severe problem.

The way you did in j2003 solves this problem, but why should it be generally nessecary to lift all ko-bans with a ko-pass in any Japanese style rule?


If there is another reason why you made the ko-pass lifting all ko-bans I would be really intered in!

[Bill Spight]

The way you did in j2003 solves this problem, but why should it be generally nessecary to lift all ko-bans with a ko-pass in any Japanese style rule?

Well, of course, nothing is necessary. However, if you believe that it is desirable to end play in an unambiguously scorable position, then that score should be the same regardless of which player plays first, with no ko ban. Thus, one of the first proposals for written Japanese go rules, Yasunaga's Go Constitution, had a three pass rule for ending play. It did not explicitly talk about lifting ko bans, but otherwise why the need for the third pass? My first set of rules ended play with the second pass by the same player in the same position, which normally would be a third pass rule. I also explicitly had passes lift ko bans.

[asura]

I did not mean passes during the game but during the determination of life and death. The rule in j1989 is that one ko-pass only lifts one specific ko-ban, but in j2003 one ko-pass lifts all ko-bans.

I don't know Yasunaga's Go Constitution, but i beleave the 3 passes are meant to avoid someone passing after the opponent has passed (e.g. because there are no neutral points to play) with an open ko. If so it wouldn't make sense if the passes don't lift a ko ban - here it would give the result whether a pass lifts one ko ban or all.

[Bill Spight]

I did not mean passes during the game but during the determination of life and death. The rule in j1989 is that one ko-pass only lifts one specific ko-ban, but in j2003 one ko-pass lifts all ko-bans.

The J89 pass for ko rule is an attempt to localize the effects of kos, but in the end does not do so for multiple kos. If you then abandon the idea that multiple kos are independent, you get pass lifting all ko bans. In my Japanese style rules I started out having pass lift all ko bans, but got some results that seemed far from the spirit of the J89 rules. So I went back to a pass for ko rule and adopted a different rule to handle global effects, the once only rule. (A player can take a ko only once in a certain position of the whole board.)