Loading field map
Hand efficiency is the study of how to choose which tile to discard in mahjong so that a player’s hand reaches a winning state as quickly as possible, on average. It is a subfield of mahjong strategy that focuses on the mathematical and probabilistic structure of tile selection, independent of the tactical concerns of reading opponents’ hands or managing defense. The central question is simple to state but surprisingly deep: given a hand of thirteen tiles and a newly drawn fourteenth, which discard leaves the hand with the fewest expected turns to completion?
The stakes are foundational. Mahjong is a game of racing—four players each try to complete a legal hand before the others. Every discard is a commitment: it reduces the set of future draws that improve the hand and, in many variants, signals information to opponents. Hand efficiency isolates the first of these costs, asking what a purely self-interested, draw-optimal player would do. It is the baseline against which all other strategic considerations—defense, tile safety, hand value, and reading opponents—are measured. A player who cannot identify the efficient discard will lose to one who can, regardless of skill in the social or defensive dimensions of the game.
The formal backbone of hand efficiency is the concept of shanten, a Japanese term meaning "number of tiles away from tenpai." Tenpai is the state where a hand needs only one more tile to win. A hand in shanten 1 needs one tile to reach tenpai; shanten 2 needs two tiles to reach tenpai, and so on. The goal of efficient play is to reduce shanten as quickly as possible, and the immediate question at each discard is which option minimizes the expected number of draws to reach tenpai, and then to win.
The key computational object is tile acceptance (or ukeire in Japanese). For a given hand configuration, acceptance is the count of distinct tiles that, when drawn, improve the hand—typically by reducing shanten by one. A discard that leaves the hand with 20 accepting tiles is better than one that leaves 12, all else being equal. But acceptance is not the whole story. Two discards may leave the same number of accepting tiles, yet differ in the quality of those tiles: some improvements lead to flexible shapes with high future acceptance, while others lead to dead ends. Hand efficiency therefore involves comparing not just immediate acceptance but the expected future acceptance after each possible improvement.
A simple example illustrates the method. Consider a hand with the tiles 2-3-4 of characters, 5-6 of bamboo, and 7-8 of dots, plus assorted unrelated tiles. The 5-6 bamboo waits on 4 or 7 bamboo; the 7-8 dots waits on 6 or 9 dots. If forced to discard one of the two-sided sequences, a player must weigh which wait has higher acceptance. Both are two-sided waits with eight tiles total, but the surrounding tiles in the hand may make one wait more valuable—for instance, if the hand already contains a 4 bamboo, the 5-6 wait is partially blocked. Hand efficiency formalizes such intuitions into a systematic comparison of counts and shapes.
A substantial portion of hand efficiency knowledge is organized around a taxonomy of tile configurations, each with known properties. The most important distinction is between blocks—groups of tiles that can form a complete set (a triplet or a sequence)—and floating tiles that are not yet part of any block. A standard hand aims to have five blocks: four complete sets and one pair, or some combination that can be rearranged into the winning structure.
The fundamental shapes are:
Beyond these basic waits, hand efficiency analyzes complex shapes: configurations where a single group of tiles can serve multiple purposes. The classic example is the sanshoku (three-color straight) pattern, where tiles like 4-5-6 in all three suits create overlapping possibilities. More important for efficiency are shapes like 4-5-6-7, which can be split as 4-5-6 plus a floating 7, or 4-5 plus 6-7, giving two two-sided waits. The 4-5-6-7 shape is famously strong because it offers multiple ways to complete a sequence while maintaining flexibility.
The taxonomy extends to isolated tiles and their value. A lone 5 is generally more valuable than a lone 1 or 9 because it can form sequences with tiles on both sides (3-4, 4-6, 6-7, etc.), whereas a 1 can only form sequences with 2-3 or 3-4. This asymmetry underlies the common heuristic that middle tiles are more efficient than terminal tiles, though the exact valuation depends on the rest of the hand.
Hand efficiency as a practical discipline is dominated by two complementary approaches. The first is a body of heuristic rules developed over decades of competitive play, particularly in Japanese riichi mahjong, where the scoring system rewards speed and the game is played with a formalized set of efficiency principles. These rules are taught to beginners and refined by experts. They include maxims like "keep the five-block structure," "prefer two-sided waits over closed waits," "discard isolated terminals before isolated middles," and "avoid redundant tiles that duplicate existing waits."
The second approach is exact computation. Because mahjong hands are finite and the tile set is small (34 distinct tiles, four copies each), it is possible to compute the exact expected number of draws to win for any given hand, under the assumption of perfect draws from a full wall. This computation is done by dynamic programming over the state space of possible hands, and it yields a precise ranking of discards. Modern software and online tools implement this calculation, and professional players use them for post-game analysis.
The relationship between the two approaches is one of mutual calibration. The heuristic rules are, in effect, compressed summaries of the exact computations. When a heuristic conflicts with the computed optimum, the exact result is authoritative—but the heuristics are valuable because they are fast, require no computation, and cover the vast majority of practical positions. The exact computation, in turn, reveals the limits of heuristics: there are hands where the "obvious" discard is suboptimal by a small margin, and these edge cases are where expert play diverges from rule-following.
A crucial limitation of the exact computation is its assumption of a full, random wall. In actual play, tiles already discarded or held by opponents are unavailable, and the wall is not random conditional on the information revealed. Hand efficiency in the strict sense ignores this information, treating all unseen tiles as equally likely. This is a deliberate idealization: it isolates the draw-optimal choice from the informational game. A player who incorporates opponent discards into the probability calculation is doing a more sophisticated form of efficiency, sometimes called information-aware efficiency, but this is a distinct practice that blends hand efficiency with reading.
The systematic study of hand efficiency emerged in Japan in the late twentieth century, alongside the codification of riichi mahjong strategy. Early strategy books focused on basic shape recognition and the five-block principle, but the field matured as players began to analyze hands quantitatively. The advent of personal computers in the 1990s made exact computation feasible, and by the 2000s, software tools could evaluate any hand in milliseconds. This shifted the field from purely heuristic to a hybrid discipline, where heuristics are tested against computation and refined accordingly.
The Japanese term ukeire (acceptance) became standard in the literature, and the concept of shanten was formalized as a countable measure. Chinese mahjong traditions, which have their own rich strategic literature, did not develop the same formal apparatus, partly because many Chinese variants have different winning conditions and scoring that make speed less uniformly dominant. The hand-efficiency framework is therefore most fully developed for riichi mahjong, though the underlying mathematics applies to any variant where the goal is to complete a standard hand of four sets and a pair.
Today, hand efficiency is a mature subfield with a clear consensus on its core methods. The exact computation is well understood and implemented in multiple open-source and commercial tools. The heuristic rules are standardized in textbooks and online guides, and they are remarkably consistent across sources—a sign that the field has converged on a stable body of knowledge.
The open questions are not about the basic mathematics but about the boundaries of the subfield. One active area is the interaction between hand efficiency and hand value: a player may choose a slower path that yields a higher-scoring hand, and the optimal choice depends on the game situation (e.g., whether the player is in last place and needs a big hand). This is sometimes called efficiency-value tradeoff, and it is a recognized extension of pure hand efficiency rather than a rival approach.
Another boundary is the integration of efficiency with defense. In riichi mahjong, a player who discards a tile that an opponent can use to win is taking a risk. Pure hand efficiency ignores this risk, but real play requires balancing speed against safety. The subfield of defense is separate, but hand efficiency provides the baseline: a player who knows the efficient discard can then decide whether to deviate for safety, and by how much.
The durable contribution of hand efficiency is its demonstration that mahjong, despite its complexity and social elements, has a rigorous mathematical core. The field does not claim to solve the game—the full game with four players, hidden hands, and scoring is far too complex for exact solution—but it solves the subproblem of draw-optimal play, and that solution is both practically useful and theoretically clean. For the educated newcomer, hand efficiency is the entry point to serious mahjong strategy: it is the part of the game that can be learned with certainty, and it provides the foundation on which all other strategic skills are built.