Whole-board seki is common in the current 5×5×5 and 7×7×7 games, and balanced komi appears to be around 1 point.
Much of 3D Go revolves around large-scale connection, enclosure, and coexistence across several layers of the board.
Board-game AI catalog
Fifty entries in one download-oriented ledger: latest packages, source branches, datasets or models, and a concise game conclusion where one is available.
No entries match this search and filter.
The game column and column titles stay visible while you scroll. Use either horizontal scrollbar.
Whole-board seki is common in the current 5×5×5 and 7×7×7 games, and balanced komi appears to be around 1 point.
Much of 3D Go revolves around large-scale connection, enclosure, and coexistence across several layers of the board.
在目前的 5×5×5 和 7×7×7 对局中,全盘双活很常见,平衡贴目大约是 1 目。
三维围棋的重点不只是局部吃子,更在于跨越多层棋盘的大范围连通、包围与共活。
The release implements Go on 5×5×5 and 7×7×7 cubic lattices. Use the packaged documentation and source for exact liberty, scoring, and komi behavior.
发布版在 5×5×5 与 7×7×7 立方格点上实现围棋;气、计分与贴目的精确定义以程序包文档和源码为准。
The engine supports boards up to 19×19, capture targets N≤5, and no pass. A player wins on reaching the configured capture target.
引擎支持最大 19×19、吃子目标 N≤5,且不能停一手;先达到设定吃子目标的一方获胜。
This is a life-and-death Go setup with seeded starting stones and a survival/kill objective. The included ReadMe and openings define the exact starting positions and winner condition.
这是带预置棋子的围棋死活变体,以做活或杀棋为目标;精确开局与胜负条件以程序包内 ReadMe 和开局文件为准。
Go rules on a hexagonal grid.
在六角格棋盘上使用围棋规则。
A move cannot be played next to the immediately previous move.
不能落在紧邻上一手的位置。
Capture Go on a hexagonal grid.
六角格棋盘上的吃子棋。
Areas or stones use different scoring weights.
不同区域或棋子使用不同的计分权重。
When a group has no liberties, the current player's stones are removed first.
棋块无气时,先移除当前落子方的棋子。
One captured stone scores four points, making captures much more valuable than in ordinary Go.
吃一子计四目,使提子价值远高于普通围棋。
A related scoring experiment can be read as no pass: the first player without a legal move loses.
同组计分实验可近似理解为禁止 pass:先没有合法着法者输。
Black can usually make a living connected group.
The current positions therefore tend to favor Black reaching this variant's winning condition.
黑方通常不难做出一块连通活棋。
因此,目前的局面整体偏向黑方达成这一变体的获胜条件。
White wins after capturing one black stone; Black wins when one fully connected group is alive. Branch notes suggest Black can often live.
白棋吃到一枚黑棋即胜;黑棋形成一块全部连通的活棋即胜。分支记录认为黑棋通常不难成活。
The desktop build offers freestyle Gomoku, exact-five/overline-does-not-win, Renju, and Caro modes; choose the matching rule preset before analysis.
桌面版提供自由五子棋、长连不胜、连珠与 Caro 模式;分析前需要选择对应规则预设。
Caro uses the dedicated rule mode included in the main Gomoku package. Consult the branch for blocked-end and overline edge cases.
Caro 使用主五子棋程序包中的专用规则模式;堵头与长连等边界情况以该分支实现为准。
I have found a 35-move first-player winning line, making Pente highly likely to be a first-player win.
The pie rule reduces the opening advantage substantially, but still does not appear to make the game fair.
This is a large-search result rather than a strict exhaustive proof.
我目前找到了一条先手 35 手取胜的主变化,因此 Pente 高度疑似先手必胜。
交换规则虽然明显缩小了先手优势,但仍没有把游戏真正拉回公平。
这是大规模搜索得到的结论,还不是严格的穷举证明。
The package supports Pente and Pente-Keryo: a player can win by making the target line or by satisfying the capture condition.
程序包支持 Pente 与 Pente-Keryo:既可以通过连成目标长度获胜,也可以通过达到吃子条件获胜。
3D Connect6 usually lasts about 70 moves, with Black winning roughly 70% of current self-play games.
Ordinary 3D Gomoku gives the first player such a direct advantage that I did not pursue it further. By contrast, 3D Connect7 and axis-only Connect5 often exceed 200 moves and look more likely to be draws.
三维六子棋通常约 70 手结束,目前的自对弈黑胜率约为 70%。
普通三维五子棋的先手胜法过于直接,因此没有继续作为重点研究;三维七子棋和仅坐标轴连五则经常超过 200 手,我更倾向于它们是和棋。
The package contains 3D Connect6, exact-six, Connect7, and axis-only Connect5 on 11³ or 13³ boards; ordinary 3D Gomoku is intentionally omitted.
程序包包含 11³ 或 13³ 棋盘上的 3D 六子棋、恰好六连、七子棋与仅轴向五连;普通 3D 五子棋被有意省略。
Some intersections are unavailable for play.
部分交叉点禁止落子。
The move completing five must land at an endpoint of the line.
完成五连的一手必须落在该线的端点。
Uses a nonstandard two-line connection structure.
采用不同于普通五子棋的双线连接结构。
After two free moves, legal points depend on the odd/even count of black stones along rows, columns, or diagonals; five wins.
前两手自由;之后合法点取决于横、竖或斜线上黑子数的奇偶性,先成五连者胜。
Scores the number of completed five-in-a-row lines instead of ending at the first one.
按形成五连的数量计分,而不是第一次五连就结束。
Combines connection and capture objectives.
结合连子与吃子两种目标。
The six target stones occupy an arithmetic progression rather than consecutive points.
目标六子位于等差数列位置,而非普通连续位置。
When only horizontal and vertical lines of four count, the first player has a fairly simple winning method.
A few selected openings remain nontrivial and are still worth studying separately.
只计算横向和纵向四连时,先手存在相当简单的必胜方法。
不过,仍有少数特定开局并不平凡,值得单独继续研究。
Only horizontal or vertical four counts. First player has a simple win, although selected openings remain nontrivial.
只有横向或纵向四连有效。先手存在简单胜法,但部分开局仍有一定复杂度。
Players alternately place stones on a hexagonal grid and try to connect their two opposite sides; draws are impossible under standard Hex rules.
双方在六角格上轮流落子,目标是连接各自对应的两条对边;标准 Hex 规则下不会和棋。
Black is highly likely to win 19×19 Connect6, probably within 109 stones (55 moves).
The current searches give Black a 92% win rate on 19×19. On 25×25 the advantage is even clearer: wins appear within about 89 stones (45 moves), with a 96% Black win rate.
Sizes up to 18 look more likely to be draws, size 20 remains unclear, and size 21 may require about 93 stones (47 moves) to decide.
These are strong model findings, though not yet a strict exhaustive proof.
19×19 六子棋中,黑方高度疑似必胜,取胜所需长度很可能是 109 子(55 手)。
目前搜索中的 19×19 黑胜率为 92%。25×25 的黑方优势更明显,胜局约需 89 子(45 手),黑胜率为 96%。
18 路及以下更像和棋,20 路暂时无法判断,21 路则可能需要约 93 子(47 手)分出胜负。
这些模型判断已经很强,但还不是严格的穷举证明。
Black places one stone first, after which each player places two stones per turn; the first unbroken line of six wins.
黑方首回合落一子,之后双方每回合各落两子;率先形成连续六子者获胜。
On 7×7 Ataxx, the first player wins about 57% of current self-play games.
The result points to a small but persistent first-player advantage rather than a heavily one-sided opening.
7×7 Ataxx 目前的自对弈先手胜率约为 57%。
这表现为轻微但稳定的先手优势,而不是非常悬殊的开局优势。
Pieces clone to an adjacent square or jump two squares; after a move, adjacent opposing pieces change color, and the larger final army wins.
棋子可复制到相邻格或跳到两格外;落子后相邻敌子会变色,终局棋子较多者获胜。
I have produced a preliminary outcome matrix for rectangular boards up to 20×20.
The overall pattern is already visible, although a small number of borderline cells may still be revised in later runs.
我已经整理出长宽不超过 20 的矩形棋盘初步胜负矩阵。
整体规律已经比较清楚,不过少数边界格子仍可能在后续计算中修正。
Players drop pieces into columns under gravity; the first horizontal, vertical, or diagonal line of four wins.
双方在重力作用下向列中落子;率先横向、纵向或斜向连成四子者获胜。
Standard Animal Chess currently draws in more than 99% of self-play games, making it look very close to a theoretical draw.
Even in the leopard–wolf–dog handicap test, draws remain above 60%; decisive advantages are unusually hard to sustain.
标准斗兽棋目前的自对弈和棋率超过 99%,已经非常接近理论和棋。
即使在豹、狼、狗让子测试中,和棋率仍超过 60%;双方都很难长期保持决定性优势。
This entry uses the standard Dou Shou Qi / Jungle rules implemented by the Dandelion interface and the linked source branch.
此条目使用 Dandelion 界面与对应源码分支实现的标准斗兽棋规则。
Current Chinese Checkers self-play is very close to a draw.
Neither side has shown a stable, substantial advantage.
中国跳棋目前的自对弈结果非常接近和棋。
先后手都没有表现出稳定而明显的优势。
The playable package implements Chinese Checkers; use its Chinese instructions for the exact player setup and move rules.
可玩程序包实现跳棋;具体人数配置与行棋规则以包内中文说明为准。
NoGo's first-player advantage decreases as the board grows but remains substantial: 87% on 9×9 and 73% on 13×13.
不围棋的先手优势会随棋盘增大而下降,但仍然相当明显:9×9 为 87%,13×13 仍有 73%。
NoGo uses ordinary Go placement, but no move may capture stones or be suicidal.
A move is illegal if it would remove an opposing group or leave the played group without liberties.
The usual losing condition is being the first player with no legal move.
不围棋沿用围棋的落子方式,但任何一手都不能提子,也不能自杀。
如果某一步会移除对方棋块,或让落下的己方棋块无气,该步就是非法着。
通常以先没有合法着法者判负。
On the standard 10×10 Amazons board, the first player wins about 70% of current self-play games.
The game therefore shows a fairly clear first-player advantage at the standard size.
标准 10×10 亚马逊棋中,目前的自对弈先手胜率约为 70%。
在标准尺寸上,先手优势已经相当明确。
Each turn moves one amazon like a chess queen, then fires an arrow from its destination to block a square permanently; the player with no move loses.
每回合先像国际象棋后一样移动一枚亚马逊,再从终点射箭永久封锁一格;无合法着法者负。
On the standard 8×8 board, the first player wins about 94% of current Breakthrough games and is highly likely to have a forced win.
At 10×10 the first-player win rate drops to 59%, producing a much more balanced game.
标准 8×8 突破棋的先手胜率约为 94%,先手高度疑似必胜。
扩大到 10×10 后,先手胜率降到 59%,局面明显更接近平衡。
Pawns move one square straight or diagonally forward and capture diagonally; reaching the opponent's back rank wins.
兵向前直走或斜走一格、斜向吃子;率先到达对方底线者获胜。
A legal move brackets one or more opposing discs in at least one direction and flips them; the player with more discs at the end wins.
合法落子必须在至少一个方向夹住对方棋子并将其翻转;终局棋子较多者获胜。
Without the tiebreaker, first-player wins, second-player wins, and draws are 21%, 2%, and 77%, making the game strongly drawish.
With the tiebreaker, the split becomes 78%, 4%, and 18%, greatly amplifying the first-player advantage.
This single rule choice fundamentally changes the balance of the game.
不使用加赛规则时,先胜、后胜、和棋分别为 21%、2%、77%,整体明显偏向和棋。
加入加赛规则后则变为 78%、4%、18%,先手优势被大幅放大。
这一项规则选择会从根本上改变游戏的平衡性。
Moves are played inside nine local tic-tac-toe boards, and each move sends the opponent to a corresponding local board; the linked release documents its tiebreaker option.
棋盘由九个局部井字棋组成,每步会把对手送到对应小盘;发布说明记录了可选的加赛判定规则。
My current conjecture is that standard Score Four is a first-player win.
For generalized Connect-N, I suspect the first player wins when N≥5 is odd, while the second player wins when N is even. This parity pattern remains a conjecture.
我目前的猜想是:标准 Score Four 先手必胜。
推广到连 N 后,我猜测奇数 N≥5 时先手必胜,偶数 N 时后手必胜;这个奇偶规律目前仍是猜想。
Score Four is three-dimensional connect four; the release also includes generalized N=5, 6, and 8 model variants.
Score Four 是三维四子棋;发布版还包含推广到 N=5、6、8 的模型变体。
Players alternate drawing one unclaimed edge. Completing a box claims it and grants another move; the player with more boxes at the end wins.
双方轮流画一条尚未占用的边;围成一个格子即可占有它并继续行棋,终局格子更多者获胜。
On a turn, move the pawn or place a wall without sealing every path. The first pawn to reach the opposite side wins.
每回合选择移动棋子或放置挡板,但不能彻底封死所有路径;率先到达对边者获胜。
Pieces move to adjacent empty points; a capture travels around at least one board loop before landing on an opposing piece. The linked reference covers variant details.
棋子通常移到相邻空点;吃子路线必须经过至少一个棋盘环线后落到对方棋子处,具体变体细节见外部参考。
Move one of your stones orthogonally onto an adjacent opposing stone and remove it. The first player with no legal move loses.
把己方棋子正交移动到相邻的对方棋子上并将其吃掉;率先没有合法着法者负。
These games reverse “complete the target line to win” into “complete the target line to lose.”
Anti-five, anti-four, and anti-three share the same core rule and differ only in target line length.
这类规则把“先连成目标长度获胜”反转为“先连成目标长度判负”。
反五、反四与反三的核心相同,只是目标连线长度不同。
Zhen Qi on boards up to size 6 is likely a draw.
On 8×8, Black appears able to win in roughly 41 moves; on 10×10, the likely Black win takes about 35 moves.
6 路及以下的震棋很可能是和棋。
8×8 中黑方疑似可在约 41 手取胜,10×10 中则疑似可在约 35 手取胜。
After placement, adjacent stones are pushed one point outward when possible; after movement resolves, a horizontal, vertical, or diagonal line of four wins.
落子后,相邻棋子会在可能时向外推一格;移动结算后,横、竖或斜向连成四子者获胜。
Same-color stones form groups through eight-neighbor connectivity. A group of at least six is removed when it is symmetric across the horizontal axis, vertical axis, or either diagonal of its bounding box.
Removing a shape of n stones scores n − 5 points. The default target is four points; the first player to reduce the remaining target to zero wins. Passing loses.
同色棋子按八邻接组成连通块。至少六子的连通块如果关于包围盒的水平轴、垂直轴或任一对角线对称,就会被移除。
移除 n 枚棋子的形状可得 n − 5 分;默认目标为 4 分,先把剩余目标降到 0 的一方获胜。Pass 判负。
The first move may use any empty point. Every later move must be within the eight-neighborhood of at least one opposing stone—the squares a chess king could reach in one step.
There is no gravity. The first player to make four horizontally, vertically, or diagonally wins.
第一手可落在任意空点;之后每一手必须位于至少一枚对方棋子的八邻域,也就是国际象棋中王一步可达的位置。
棋子不受重力影响;先在横、竖或斜线形成四连者获胜。
On a turn, either place one of your stones on an empty point, or move an opposing stone like a chess queen along an unobstructed line; its original point is then replaced with your stone.
When either color makes five, that color loses. Passing loses; a full board with no five is a draw.
每回合可以在空点放一枚己方棋子,也可以选中一枚对方棋子,让它沿横、竖或斜线移动到路径无阻的空点;原位置随后换成己方棋子。
任一颜色形成五连时,该颜色判负。Pass 判负;棋盘填满且没有五连则和棋。
Each turn places one own stone and optionally one limited neutral ban point. After 16 stones each, relation patterns score; fewer neutral bans break a tie.
每回合放一枚己方棋子,并可放一枚有配额的中立禁点;双方各 16 子后按关系模式计分,同分时中立禁点较少者胜。
This experiment follows standard Xiangqi rules; use the linked reference for piece movement, check, and palace or river restrictions.
此实验采用标准中国象棋规则;棋子走法、将军以及九宫和河界限制见外部参考。
This early experiment follows standard chess rules; use the linked reference for the complete move and checkmate rules.
这次早期实验采用标准国际象棋规则;完整走法与将死规则见外部参考。