Loading field map
The yose is the final phase of a game of Go, the sequence of play that begins when the large-scale territorial frameworks and middle-game fights have been resolved and the board settles into a series of smaller, more local exchanges. The term itself is Japanese, literally meaning "to finish" or "to tighten," and it refers both to the remaining moves on the board and to the disciplined technique of playing them correctly. In the broader arc of a Go game—opening, middle game, and endgame—the yose is often the least glamorous, but it is rarely the least important. Professional games are frequently decided by margins of a few points, and those points are won or lost in the endgame. For a learner, the yose is the phase where abstract positional judgment gives way to concrete calculation, and where the game becomes, in a limited but real sense, a problem in arithmetic.
The fundamental question of the yose is deceptively simple: given a set of remaining moves, each of which gains some number of points for the player who takes it, in what order should they be played? The difficulty arises because the moves are not independent. A move in the corner may threaten a follow-up that is worth more than the move itself; a move on the side may be urgent because it prevents the opponent from playing there first; and the value of a move can change depending on what has already been played elsewhere. The yose player must therefore evaluate each candidate move not only for its immediate point gain but also for its size relative to other moves, its urgency, and its effect on the value of subsequent moves.
The basic unit of analysis is the point, the smallest unit of territory or captured stone. A move that secures two points of territory in a corner is said to be worth two points. But this simple counting is complicated by the fact that most endgame moves are sente or gote. A sente move is one that forces a response; the player who makes it retains the initiative and can then play another move elsewhere. A gote move is one that does not force a response; the player who makes it yields the initiative to the opponent. The distinction is crucial because a small sente move may be worth more in practice than a larger gote move, since the sente move costs nothing in terms of initiative. The standard way to compare moves is to calculate their value: the total number of points at stake, divided by the number of moves required to secure them. A move worth six points in gote, for example, is roughly equivalent to a move worth three points in sente, because the sente move uses only one move while the gote move uses two (one for each player).
This arithmetic is the backbone of the yose, but it is only the beginning. The true difficulty lies in the fact that the value of a move is not always fixed. A move that appears to be worth five points may, if played at the wrong time, allow the opponent to respond in a way that reduces its value or creates a larger threat elsewhere. The endgame is therefore a dynamic process, not a static list of independent values. The skilled player must constantly re-evaluate the board, updating the values of remaining moves as the situation changes.
For most of Go's long history, the yose was treated as a matter of intuition and experience. Players learned through practice which moves were large and which were small, and the endgame was seen as a less interesting, more mechanical version of the middle game. This began to change in the twentieth century, particularly in Japan, where professional players and theorists started to analyze the endgame more systematically. The key insight was that the endgame could be treated as a combinatorial problem: a set of independent or weakly interacting local positions, each with a well-defined value, and the task of finding the optimal order of play.
The most influential figure in this development was the Japanese professional player and writer Otake Hideo, whose books on the endgame, published in the mid-twentieth century, introduced a generation of players to the idea of counting moves in terms of their point values and to the distinction between sente and gote. Otake's work was not entirely original—earlier players had understood these concepts informally—but he codified them into a teachable system. His approach was essentially practical: he showed how to calculate the value of common endgame shapes and how to use those values to decide the order of play.
A more rigorous, mathematical treatment emerged later, largely through the work of the British mathematician John Conway and his collaborators, who developed a theory of combinatorial games in the 1970s. Conway's theory, originally developed for the analysis of games like Hackenbush and Domineering, was applied to Go endgames by the American mathematician Elwyn Berlekamp and others. This approach, sometimes called combinatorial game theory (CGT), treats each local endgame position as a game in its own right, with a numerical value that can be added to the values of other positions to determine the overall outcome. The theory is elegant and powerful, but it has significant limitations. It works well for positions that are truly independent, but real Go endgames often involve interactions between positions, and the theory becomes unwieldy when those interactions are significant. Moreover, the theory assumes perfect play, which is rarely achievable in practice. For these reasons, CGT has remained a specialized tool, more useful for analyzing artificial or simplified positions than for guiding play in real games.
Despite the mathematical sophistication of CGT, the practical yose is dominated by two complementary approaches: counting and reading. Counting is the attempt to assign numerical values to moves and to use those values to determine the order of play. Reading is the attempt to see, move by move, what will actually happen on the board. The two are not rivals; they are tools used together. A player who only counts may miss a subtle sequence that changes the value of a move; a player who only reads may waste time on variations that could have been dismissed by a simple calculation. The skilled yose player moves fluidly between the two, using counting to narrow the field of candidates and reading to verify the details.
The counting approach is most useful in the early endgame, when the board is still relatively open and there are many moves of comparable size. The player who can quickly estimate the values of the remaining moves and identify the largest ones has a significant advantage. The reading approach becomes more important as the board fills in and the number of moves shrinks. In the final stages of the yose, when only a handful of moves remain, the game becomes a pure exercise in reading: the player must see the exact sequence of play, including all possible responses, to determine the correct order.
There is also a third, more subtle skill that is often overlooked: awareness of the whole board. The yose is not played in a vacuum. The value of a move in one corner may depend on the presence of a weakness in another part of the board, or on the number of ko threats available. A move that appears small may be urgent because it prevents the opponent from creating a large threat elsewhere. This holistic awareness is difficult to teach and is often what separates a strong amateur from a professional. It is also the reason why the yose cannot be reduced to a simple algorithm: the interactions between local positions are too complex for any fixed rule to capture.
The yose is complicated by several special situations that do not fit neatly into the counting framework. The most important of these is the ko, a situation in which the board position repeats itself after a single move and a capture, and which is governed by a rule forbidding immediate recapture. Ko fights can be large and complex, and they often arise in the endgame. The value of a ko is not simply the number of points at stake; it also depends on the number of ko threats available to each player—moves elsewhere on the board that are large enough to force a response, allowing the player to retake the ko. The presence of a ko can therefore change the value of moves that are not directly involved in the ko itself, since a move that creates a ko threat may be worth more than its face value.
Another complication is the sente-gote distinction itself. In practice, the distinction is not always clear-cut. A move may be sente in one context and gote in another, depending on the size of the threats it creates. A player may also choose to play a move that is technically gote because it is kikashi—a forcing move that improves the player's position even if the opponent responds. These subtleties mean that the simple arithmetic of the yose is always an approximation, and the skilled player must be aware of when the approximation breaks down.
In the contemporary game, the yose has been transformed by the advent of computer Go. The strongest programs, based on deep neural networks and Monte Carlo tree search, play the endgame with a precision that exceeds that of any human. They do not use the traditional concepts of sente, gote, or counting in the way that humans do; instead, they evaluate positions through massive search and pattern recognition. This has led to a curious situation: the computer's play in the endgame is often surprising to human observers, and sometimes appears to violate traditional principles. Yet the computer's results are consistently better, which has forced human players to re-examine their assumptions.
The impact on human play has been significant but not revolutionary. Professional players have studied computer endgame play and have adopted some of its insights, particularly in the area of move ordering and in the handling of ko fights. But the fundamental concepts of the yose—counting, reading, sente, gote—remain the tools that humans use, because they are the tools that humans can use. The computer's advantage lies in its ability to calculate far more deeply and accurately than any human, not in a fundamentally different understanding of the game. The yose remains, for human players, a discipline of careful arithmetic and precise reading, a final test of patience and accuracy after the chaos of the opening and the middle game.
For the learner, the yose is the most accessible part of Go to improve at quickly. Unlike the opening, which requires broad judgment, or the middle game, which requires tactical vision, the endgame rewards simple, careful practice. The player who learns to count the values of common shapes, to distinguish sente from gote, and to read the final sequences accurately will gain points in almost every game. The yose is where games are won and lost, and it is also where the game becomes, for a brief time, a pure and satisfying puzzle.