Renaissance Rules
This project is about modeling the complete game as played in China, Japan and Korea, including triple ko draws and (almost) unrestricted ko play.
Table of contents |
Summary
AGA/NZ were the first simple and logical rules, but instead of formalizing the existing game, they described a reduced variant by removing various legal moves, and long cycle draw on repetition that is prevalent in Asia. At the same time, NZ rules reintroduced score draws with integer komi 7.
These rules avoid that reduction, with an equally simple and logical model of gameplay that handles all repetitive situations. Usable with both area and territory scoring (the latter with pass stone playout):
- basic ko with a threefold repetition rule
(draw/loss/win depending on stones captured in the loop)
- minimal-superko:
a board play must not recreate the position of the last successive two passes
- play stops on three successive passes
In other words: two passes block later repetition of the position, three passes stop play completely.
Minimal-superko is easier to apply than normal superko since at most one position needs to be remembered, not whole history. It is also less intrusive since its effects are limited to the necessary minimum.
NOTE: The above refers to V2 of this project, please see [toc17] for V3 plans. Also, early versions didn't have real territory mode, but the intent is to add that as well based on something ilke this. So a future V3 would probably be as close to common minimal rules as possible:
- basic ko with a prisoner-aware long cycle rule
(draw/loss/win depending on stones captured in the loop)
- normal two pass stops
- optional moonshine ko rule (either minimal superko, or superko since the last pass)
- support for both area and territory scoring, as well as territory + pass stone playout hybrid
Alternative for AGA/NZ
Since AGA/NZ rules only target area scoring, they don't need three pass stops and could consider an alternative way of supporting draw on repetition. Most of their rules could remain untouched with only a slight refinement:
- for main phase: basic ko with threefold repetition
(draw/loss/win depending on stones captured in the loop)
- superko only applies in dispute phases, after first resumption
Having this at least as an option would make those rules (mostly) compatible to the Chinese game, ease the applicability problems of superko, and achieve better rulings - for negligible costs.
Comments
For rules following current Asian practice of draw on repetition, the first problem is how to differentiate between various types of cycles - most importantly triple ko and moonshine life. Area scoring has a further weakness in unbalanced cycles (repetition fueled by one side losing more stones). Hybrid scoring (territory + pass stone playout) faces another problem as it is only safe after three passes - which in turn creates further repetitive potential (eg. double ko seki).
Three passes were proposed by various authors before (Yasunaga and others). One motivation behind them is to get rid of technical/unintended stops, where the first pass was not real, only forced by a ko ban. Three passes are also how Spight rules usually work in practice.
This is particularly important for territory scoring, to stop main play in a ko-wise complete position, before switching to pass stones. Two pass stops could allow a player to force a stop with an open ko (using excess threats), and only connect it in the pass stone phase, for one point scoring error (like under Lasker-Maas rules). Further care is needed to switch without forcing dame fill in the main game phase already (while the players may not yet expect a dispute), preserving an advantage of territory scoring.
Although three pass stops are not strictly necessary for this model (threefold repetition and minimal-superko work equally well with two pass stops and more frequent resumptions), it is still the simplest solution for the switching problem (and a nice complement for the two pass nature of minimal-superko).
Principles
Go rules design can be a complex topic where even the objectives are questioned sometimes. The following principles can serve as guidelines:
The basic rules of go are undisputed. All rulesets agree on strings, capture and immediate basic ko recapture prohibition.
Moves that are legal according to these basics must not be prohibited without good reasons. Unnecessary restrictions are as bad as incorrect rulings.
The status and outcome of a position must depend on play - normal play with basic rules. Any differences to this must be backed with good reasons (like moonshine life).
If L/D is assigned by hypothetical play (ie. pure territory scoring, without pass stone playout), such hypothetical play also has to remain closely related to what could happen if normal play would continue.
The area and the territory game must stay reasonably close to each other. No significant behavior differences are tolerable, beyond minor or unavoidable things like last dame or scoring itself.
Comparing the two may also identify weaknesses in each.
4. Conceptual simplicity > textual conciseness
Rules need to be simple - but conciseness in itself does not imply simplicity.
Conceptually simple rules (like the above mentioned basic go rules) can always be written in a reasonably concise form. But the opposite is not true: textually concise rules can still be very complicated conceptually. A short sentence can hide references to very complex logical concepts.
5. Keep the main game phase clean
With territory scoring* it is possible to play the first game phase with just the undisputed basic rules - the capture rule and the basic ko rule, until the first stop on two passes.
Because of this - and because most players are not aware of any rules beyond those basics anyway - extra rule inventions better not touch this phase and only apply later, if there is a dispute (so never in practice).
Draft
(First draft - trivial definitions omitted, just showing the logical units and how they work together.)
- 1. Basic rules and trivial definitions (board, intersections, move = board play or pass, strings, liberties, capture - with [3] suicide allowable, resign, handicap)
- 2. Basic ko (if a single stone captures a single stone, it cannot be recaptured immediately on the next move).
- 3. Threefold repetition (draw/loss/win depending on stones captured in the loop). Applied for board plays only and thoughout all phases together.
- 4. Minimal-superko: a board play must not recreate the position of the last successive two passes.
- 5. Three successive passes stop the game and ends a game phase. Resumption from a stop is requestable, starting a new game phase, with intact move order, for a finite times. (First phase is "main game", later phases "dispute phases".)
AREA MODE:
- A.6. A player's score is the number of intersections controlled (occupied or surrounded by his live stones), plus W komi.
- A.7. In a stopped position dead strings are what their owner acknowledges, other strings alive.
TERRITORY MODE WITH PASS STONE PLAYOUT:
- B.6. A player's score is the number of intersections in his territory, plus earlier prisoners taken, plus opponent's stones in his territory ("dead stones"), plus W komi.
- B.7. In a stopped position dead strings are what their owner acknowledges, other strings alive. A player's territory are the intersections that are empty or contain a dead stone of the opponent, and are surrounded by his live stones.
- B.8. After the first dispute phase a player who passes gives a stone to the opponent as prisoner.
- B.9. After the first dispute phase the number of moves played by the players must be equal. The stopping passes of each later phase are complemented by an obligatory extra pass of the next player if necessary.
TERRITORY MODE WITH HYPOTHETICAL PLAY:
- C.6. A player's score is the number of intersections in his territory, plus earlier prisoners taken, plus opponent's stones in his territory ("dead stones"), plus W komi.
- C.7. {---}
Examples
1. 1-eye-flaw (dead ko)
captures in the ko, forcing to pass. After pass, would like to recapture (with or without resumption). This is a 2-pass non-perpetual repetition cycle. Note that basic ko only forbids immediate recapture, not later recapture.
Desired outcome: B is capturable with the above basic legal sequence.
RR: Neither threefold repetition nor minimal-superko intervenes here. The "repeating" position is not the one the last two passes occurred in. Even if B waits until two passes occur in the start position (like pass, pass, takes ko, pass, pass, recaptures ko) W is free to recapture because of / passes.
2. Double ko seki
A player may try to use this position for reaching threefold repetition or preventing a stop. There is a possible 2-pass half-cycle repetition here: flips a ko, flips the other, pass, pass, (resumption if necessary), flips a ko etc.
Desired outcome: Seki. The position is not enough for draw on repetition, and a stop is forceable.
RR: Because of / passes, after pass can deviate and capture the double ko seki by flipping in it preemptively. would need to restore the position to that of / and is forbidden. This is the same behavior as normal superko without passes lifting bans. was a mistake.
3. Triple ko
Both players are forced to play in the cycle, to avoid getting captured. This is a no-pass perpetual repetition cycle.
Desired outcome: Draw on repetition.
Notable alternative: Although undesired in the context of these rules, blocking the cycle (forcing a player to play away) is not necessarily incorrect here either, since that could mean a meaningful ko fight.
RR: Threefold repetition, with both players capturing the same number of stones so the result is draw. Minimal-superko can have no effect here since the position is not stable so couldn't have seen two successive passes.
4. Moonshine life
W tries to capture starting at 'a'. B resists by flipping the double ko 'b/c', then restoring 'a'. There are two possible cycles here: If W also uses the double ko there is a one-sided no-pass perpetual cycle. If W passes after the above sequence, there is a 2-pass (half-)cycle.
Desired outcome: B is capturable.
RR: W waits until play stops, then starts the capture in dispute. After the above sequence, W flips the double ko himself. B cannot answer since that would recreate the initial position (which both players passed in when play stopped), so B is captured.
If B gets to play first in resumption, he can try to play elsewhere to change the board (to avoid getting captured like above). W passes in response, until B runs out of moves and passes himself. At that moment both players passed in the position, so W can start the above capture. If B tries to use the double ko instead of moves elsewhere (for infinite board-changing potential), he gets captured as seen in the double ko seki example above.
5. SendingThree-ReturningOne
Assuming neither side wants to start the ko, B may try to use this position to reach draw on repetition or prevent a stop. There is a possible 1+1pass cycle here: at 'a', pass, at 'b', at 'c', restores the start position with the marked move, pass.
Desired outcome: Seki. The position is not enough for draw on repetition, and a stop is forceable.
RR: Repeating the above sequence leads to threefold repetition. But this means a loss for B since W captures two more stones with each cycle. Forcing the sequence was a mistake.
Minimal-superko has no effect here, but B could also wait for play to stop before starting repetition. This means starting from either the shown position or after 'a' - these are the two possible stable states where play could have stopped in with successive passes. In this case, if B starts from the shown position he will not be able to recapture after 'c' and will be captured. If he starts from the position after 'a', he will not be able to play more than one cycle.
6. 0-sided endgame ko
(Francisco Criado & René Martínez - see also an earlier, more complex example)
Mannenko is a common shape with a one-sided ko (only fillable by one side). Less common shapes can have kos that neither side can fill safely. Such ko remains unresolved forever - but every capture in it still affects the score.
If the game is close enough that the side with the last capture in the ko wins, this leads to a 2-pass half-cycle: takes ko, pass, pass (cannot fill), takes ko (with or without resumption), pass, pass (cannot fill), etc. The winner changes back and forth with each cycle.
Desired outcome: draw on repetition.
Note that there is no acceptable alternative here (unlike in the triple ko example above). Breaking the cycle would not lead to a meaningful ko fight (since still neither side could win the ko), it would just degenerate the game into NoPassGo. And stopping play after an arbitrary number of resumptions or encores would just randomize the winner.
RR: Draw on threefold repetition. Minimal-superko never forbids ko recapture here. For example is always valid because of / passes.
7. Molasses ko
In this position B, moving first, needs to exchange moves a-b-c-d. If he tenukis or passes instead, can respond at 'b', taking first in a direct ko for killing B. After the a-b-c-d sequence, gets to tenuki or pass once. After that, faces an identical position (second diagram) and is similarly forced to exchange f-g-h-i to avoid capture, and gets his free move at . This completes the second 1-pass half-cycle, returning to the first shape with B to move again.
Since it is never safe to pass after the opponent's pass, the cycle will continue even after running out of normal moves elsewhere, with only passes remaining for and .
Desired outcome: draw on repetition (two successive passes can not occur).
(If B is otherwise ahead, he may try to pass after pass in the first position, to force scoring. This is only possible with two pass stops, and only under rulesets that allow resumptions with changed move order (so B could hope to avoid capture by getting the first move in spite of his pass if W requests resumption). This essentially means Japanese rules - risking a "both lose" situation, with an effective killing move left on the board.)
RR: Threefold repetition, with equal captures so the result is draw. Minimal-superko has no effect without two successive passes, and the second pass is not possible since it would always let the opponent go for direct ko.
8. Bent4 with double ko seki
On area vs territory
Area and territory scoring are both widely used (in Chinese and in Japanese/Korean rules) and will remain so. Both has significant advantages and disadvantages, and they represent an important duality in Go. Nearly all rule issues are related to this - or to earlier inventions trying to fix such.
1. L/D disputes
If the players cannot agree which stones are alive, play can simply continue under area scoring. Territory scoring, on the other hand, has to revert the board and score the original position after a dispute played out, to avoid score drifts.
This is a significant flaw of territory scoring. However, such disputes are almost exclusive to beginner games, so this has little practical relevance. More important is the theoretical difficulties of string statuses (some examples here).
2. Ko and unbalanced repetition
Using just the simple ko rule, territory scoring has no problem with any repetition in main phase - infinite sequences work correctly whether the rules mention them or not (in a theoretical sense - only practical convenience is improved by clarification).
Area scoring, on the other hand, needs some extra rule in main phase already*, to make prisoner-unbalanced repetition illegal or disadvantageous. This is a significant flaw of area scoring, which makes area ko rules more complex in theory. However, territory scoring can also benefit from a robust prisoner-aware repetition rule (even as theoretical necessity in hypothetical play).
Unifying ko is not easy since ko rules are closely tied to the scoring method, and only a few combinations are viable. Normally simple ko is used for territory scoring (no superko since passes must remain usable as ko threats there, and would also be superfluous), and either Chinese ko (distinguishing types of repetition) or superko for area scoring (no simple ko because of unbalanced repetition). A promising common denominator - a logical simplification of Chinese ko - is the above mentioned prisoner-aware threefold repetition rule (with further refinement for ML).
3. Score accuracy
Area scores lose 1 bit of resolution compared to territory scores, and can normally change in 2 pts steps only (B+6 and B+7 territory both usually come out as B+7 area, masking the smallest bit). This is a significant flaw of area scoring, which:
- Awards 1 extra point for B 50% of time, as a passive gift from playing 1 more stone than W. Played stones don't directly increase territory scores, but are automatically a point in area.
- Allows B to make 1 point mistakes without affecting the score. Depending on remaining dame parity, he may choose a competitively worse move in yose (or even an unnecessary defensive move in territory, worse than passing) and still end up with the same area score.
- Makes non-integer komi less balanced for bots and strong players. The fractional (.5) part treats draws as wins for a side, which gives double advantage under area scoring.
4. Territory in seki
Area scoring has no particular problem with sekis. Territory scoring, on the other hand, needs an extra rule to exclude them from scoring, to avoid destabilization in certain positions (more here). This is a minor flaw of territory scoring.
5. Moonshine life
ML affects both scoring methods but a bit differently. Area scoring needs ko rules that make it capturable in actual play. This is the ideal approach for unifying the two, but territory scoring has another option (some kind of local view in hypothetical play may also handle the problem).
6. Hybrids
The way the two scoring methods work around each other's flaws inspired ideas that try to borrow advantages from the other. For area scoring: button go, attempting to fix the lost bit of area scores. For territory scoring: combined territory+area scoring (territory main phase followed by area-like dispute phase with pass stones).
Evolution
V1 - threefold repetition, encore-only superko
These ideas started with the above AGA/NZ alternative: a small fix for most problems of those rules - a Chinese-compatible ruleset with correct long cycle handling.
V2 - threefold repetition, minimal superko
Replacing encore-only superko with a simpler, more correct ko rule, eliminating some unwanted side effects of the former (like bent4 + double ko seki), aiming to affect moonshine kos only. Also introduced territory mode which in turn needed three pass stops.
Possible changes may worth considering
First, I think it would be better to drop three pass stops, and have the same effect (safe switch to pass stones, w/o open ko problem) achieved other ways, like switch if an encore starts with a pass, or similar. This would have a few advantages:
- Keep main game phase intact, with its well known two pass stops
- Make minimal superko sound nicer (no recreation of the encore start position)
- Make rules more modular (let double ko seki cause the usual periodic stops, for example). This could theoretically allow minimal superko to be replaced or removed in simplified versions, since then it is not relied on to prevent double ko looping.
Second, it seems reasonable (for play-out rules) to at least consider an alternative without any superko - just normal ko with a prisoner aware threefold repetition rule. This would accept moonshine life alive - in exchange for an even simpler, more natural and robust ruleset. These are, basically, the undisputed minimal rules of go - the things all players know and accept.
ML is normally uncapturable in actual play. Both Japanese and Chinese rules agree on this, only declaring it dead *after* normal/main play (Chinese rules: 1, later posts in 2, and dia3 comment2 in 3). There were periods - even recently - when ML was considered alive ( link), and I think there is logic in both opinions. Current Korean rules allow local ML and are not afraid of the one-sided draw on repetition potential. And introducing a specific moonshine rule, be it minimal superko or something else, is never completely safe or simple enough.
The ML question is not the only problem with a simplified superko-free version though, it would also need extra care about resumption rules and threefold repetition. Without relying on minimal superko, it would not be easy to allow 0-sided kos to draw on repetition despite spanning stops, while not allow this for double ko sekis.
Another minor change possibility is for AGA/NZ alternative: allow passes lifting bans (similarly to Spight rules). This is not normally possible for superko rules (reason), but is possible with the (prisoner aware) threefold repetition rule. This could even allow a simplification: instead of explicitly specifying superko for encores only, it could forbid recreation of only those positions that occurred since the last pass (so no real effect in main game). The downside is this would also need revised stop/resumption rules in those rulesets (due to the extra freedom of possible moves around stops, double ko seki etc).
Discussion
Feel free to discuss here or on L19 (more about reasons there).