Riichi mahjong strategy

Fukuchi Makoto

translated by anon

Introduction

This is a translation of a html book by Fukuchi-pro, author of 現代麻雀技術論 (Modern Mahjong

Strategy Theory). Since Riichi Book 2 by Daina Chiba probably never, I started translating this one,

since it’s extremely comprehensive and detailed and has both abstract theory and many examples.

It’s by no means a word-for-word translation: I’ve added examples to make some points clearer,

and omitted some passages that were redundant or unclear. If you can into moon runes, denitely

read the original at http://yabejp.web.fc2.com/mahjong/tactics.html.

So why this book and not some other one by a more famous contemporary pro? There are

several reasons. First of all, this is the book I personally used to get into tokujou. It’s about 12

years old, and some of the meta concepts have changed since then, but it’s still a good base. Second,

I think it lls an important gap between books for beginners and books for mahjong addicts who

already know the standard plays. You need to learn the standard plays somewhere. But if you’re

new to mahjong and haven’t read RB1 yet, go read that one rst, because it introduces many

important concepts in a more accessible way.

Who is this book for? It’s for people who want to git gud, but can’t into moon. If you don’t

care about gitting gud, you’ll probably nd this book dry and boring. It’s also not for brainlets:

there are many passages where you have to think hard, and if anything I’ve cut down on the

explanations instead of expanding. But don’t worry: even if you understand only half, you’ll notice

a big dierence once you start playing. You just start thinking about the tiles dierently after a

while.

The book is heavily focused on tile eciency, with a highly digital (analytical) way of thinking.

All of the later material about push-fold judgment and discard reading is based on good tile e-

ciency. People often claim they know tile eciency, yet they can’t solve simple WWYD problems.

I’m a Tenhou 6d and among the higher IQ gaijin players, but I still have trouble with (say) G.

Uzaku’s problems. Tile eciency is hard.

I’ll try to translate roughly 20 lmao more like 10 pages a week, we’ll see how it goes. Just (you)

me in the thread or PM me on IRC (Sjaalman) if you have a question or opinions about this shit.

Desu I just want to help everyone git gud, because I’m a tryhard and I think it’s more fun that

way.

t. attention whore

Contents

Contents 1

0.1 Outline of mahjong skills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

0.2 How to get good . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1 Tile eciency 9

Principles of tile eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.1 Tile logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Discard choice method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Isolated tiles and simple joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Comparing pairs and bad shape joints . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Complex joints I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Complex joints II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Ultra bad shape joints and furiten joints . . . . . . . . . . . . . . . . . . . . . . . . . 19

Comparing components of the same class . . . . . . . . . . . . . . . . . . . . . . . . 20

Penchan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Outer kanchan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Inner kanchan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Ryanmen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Pair dropping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Isolated 1 and 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

Isolated 2 and 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Isolated 3 to 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Joint dropping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Comparing simple and complex joints . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Comparing isolated tiles and complex joints . . . . . . . . . . . . . . . . . . . . . . . 40

Complex groups (many-sided waits) . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

1.2 Head start tenpai judgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Understanding tenpai eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Head start riichi judgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Win rate maximizing damaten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Defensive damaten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Wait improvement judgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

Wait choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

1.3 Calling eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

Calling tile logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2 CONTENTS

Calling judgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

Planning yaku for calls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

How to make open yaku . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Principles of calling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Dierences from closed tile logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

1.4 Yaku composition techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Principles of yaku and dora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Tanyao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

Pinfu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

Iipeikou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

Yakuhai pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Single yakuhai . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Sanshoku . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

Ittsuu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

Toitoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Chiitoitsu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Single suit hands (honitsu and chinitsu) . . . . . . . . . . . . . . . . . . . . . . . . . 78

Chanta, junchan, honroutou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

Sanankou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Kokushi musou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Nagashi mangan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

Dora in closed hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

Dora in open hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

Atoatozuke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

Formal tenpai . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

1.5 Iishanten eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

The importance of being rst to tenpai . . . . . . . . . . . . . . . . . . . . . . . . . . 90

Classication of iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

Extra tile shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Perfect iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

Extra tile versus perfect iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

Understanding tenpai chance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Headless iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Extra tile versus headless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

Perfect versus headless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Sticky iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

Perfect versus sticky . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

Returning to 2-shanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

2 Push-fold judgment 109

2.1 Dealing with non-tenpai opponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

Choking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

Dealing with getting choked . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

Discarding good tiles rst . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

Early game defense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

2.2 Folding techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

CONTENTS 3

Betaori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

Eectiveness of wait reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Coping with having no safe tiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

2.3 Dealing with riichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

The average score of riichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

When in tenpai . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Choosing the wait against riichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

Dealing with 2 riichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

What to do in borderline situations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

When in iishanten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

2.4 Dealing with open hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Characteristics of open hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Tenpai estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

Defending against specic yaku . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

2.5 Supplement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

Push-fold judgment just before a draw . . . . . . . . . . . . . . . . . . . . . . . . . . 123

Rolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

Ippatsu disruption and haitei shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3 Reading 127

Principles of reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

3.1 Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

The 1/18 rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

Honor waits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

3.2 Discard reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Types of discard information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Sobaten riichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Gyakugiri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Call reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

Dora discarding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Joint dropping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Pair dropping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

Yaku reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

Sobaten tanki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Chiitoitsu reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Eective wait reading situations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Hand progress reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

3.3 Wall reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

Advantages of wall reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

Tile group reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Suit reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Reex reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

3.4 Supplement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

Disguise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

Gyakugiri II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Lag reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

4 CONTENTS

Tile arrangement reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

Person reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

4 Point situation judgment 147

Principles of point situation judgment . . . . . . . . . . . . . . . . . . . . . . . . . . 147

How to think about expected placement . . . . . . . . . . . . . . . . . . . . . . . . . 148

Scoring techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

How to ght in all last . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Declining wins in all last . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

Speed maximization techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

When to attack fast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Calling when a comeback is uncertain . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Push-fold judgment in all last . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

Push-fold judgment one hand before all last . . . . . . . . . . . . . . . . . . . . . . . 153

How to make mangan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

What to do when far behind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

How to play when a player is close to shadowrealm . . . . . . . . . . . . . . . . . . . 155

Riichi sticks and honba . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

When to call kan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

0.1. OUTLINE OF MAHJONG SKILLS 5

0.1 Outline of mahjong skills

“The nature of mahjong is repetition of choice and lottery.”

—Kihara Kōichi

In these lotteries, the will of the players plays no role. That is to say, they are completely

random.

Through choices, it’s possible to make the lottery more favorable. Only

there is there space for mahjong skill to enter. Afterwards, one can only

wait for the result of the lottery.

Now, what kind of skills are required to win at mahjong (maximize expected income per han-

chan)? We’ll categorize them as below, and consider the optimal strategy for each in turn.

(1) Skills directly related to discard decisions in themselves

– Tile eciency (includes score eciency, yaku composition techniques, whether to call riichi

with a head start, calling judgment)

The skill to maximize (win rate × average score) + (expenditures when we don’t win)

– Folding

The skill to minimize (deal-in rate × average deal-in score) + (opponent tsumo rate ×average

opponent tsumo score)

– Choking and rolling

So to speak, the skill of tile eciency when considering opponents’ interaction with discarded tiles.

(When they’re not in tenpai for choking and when they are for rolling.) A kind of in-between

strategy between pure eciency and folding.

(2) Push-fold judgment (including riichi and calling judgment when an opponent is

in tenpai)

Judging which of the three strategies in (1) is best.

(3) Point situation assessment

(1) and (2) essentially aim at maximizing expected income from a single round. But in mahjong,

nal placement is also important, making round expected value 6= hanchan expected value. These

skills complement this disparity.

(4) Reading skills

Mahjong is a game of imperfect information. While (1), (2) and (3) tend to be based on known

information, these skills aim to complement them with deduced unknown information to increase

their accuracy.

– Wall reading

Assists with tile eciency in (1).

– Wait reading

Assists with (2).

– Yaku and hand progress reading

Assists with (2).

(5) Improvement methodology

The four previous skills are used in actual play. This skill is about how and how much to practice

them and how to use them successfully in practice.

6 CONTENTS

0.2 How to get good

• Knowing strategies is important, but being able to successfully put them into practice is just

as important.

• The single most important strategy is deciding whether to attack or to defend.

• Beginners should focus on either completely attacking or completely folding depending on

their push-fold decision. Intermediate strategies are less important to master and much more

dicult.

• It’s important to use all available information (own hand, discards, dora, opponent’s calls,

point situation, round etc.), but this information is not all equally important.

• Reading hidden information should only be attempted by players who already understand the

rest.

• We should always play with a concrete goal in mind. Goal ⇒ push-fold decision ⇒

discard.

• There are many strategies, but it’s best to learn the most important ones rst.

• The most important strategies other than push-fold judgment are tile eciency of good hands,

whether to call riichi, how to fold, how to deal with opponents’ calls, how to evaluate score

dierences.

• It’s important to practice in order to make less mistakes, but it’s impossible to never make

any.

• In the same situation, be consistent and play the same move.

• It’s good to work on skills we’re good at because it’s easier to get motivated.

• But it’s better to work on skills we’re bad at because it will improve our play more.

• To nd out what we’re bad at, we need to use long-term data.

• Watching strong players play can be good, but only if we focus on particular points that we’re

bad at.

• Once we know how to overcome a weakness, the best way to do it is to play many games.

• The combination of real play and study is the fastest way to learn.

• Don’t play dierently just because we’re on a losing streak.

• Always review the paifu after a game.

• When running out of time too often, we should think about what to discard during opponents’

turns.

• When making many careless blunders, we should think longer during our own turn.

0.2. HOW TO GET GOOD 7

• We should remember that mahjong is a game of chance and not get mad at unlikely

events.

• If we still get tilted and start losing due to tilt, we should stop playing.

• When playing, concentrate on the game.

• Don’t play when hungry, sleepy, sick, stressed or overworked if it means we can’t concentrate.

• Take good care of our health and condition to be able to play in tourneys at inconvenient

times.

This list is a short-form summary of chapter 5

Chapter 1

Tile eciency

Principles of tile eciency

We will rst discuss the most important things we need to pin down.

By tile eciency, we mean the maximization of expected value at rst without explicitly consid-

ering opponents attacking. (The decision what to do when opponents attack is called push-fold, the

techniques for dealing tiles when opponents are attacking are called choking, rolling and folding.)

Maximizing win rate

The expected value for a round is

win rate × score + point expenditure when not winning

Accordingly, we should maximize win rate, since it will increase both terms of the equation.

What this is means in practice is, when aiming for a big hand, don’t overlook the compatible

cheap hands.

Example erssffjjk,.5567

With this kind of hand, we want to stack the dragons and draw more pinzu, aiming for honitsu.

Accordingly, we should break up ,.. However, breaking up the er rst is unacceptable,

because it’s a better shape than ,. and the wt acceptance enables us to make a quick cheap

hand.

Leaving room for choice is usually not advantageous

Usually, when there is room for a choice, we simply want to not overthink it and take the best

move. It’s not necessary to be perfectly consistent, but for each situation we want to know the

basic move and if needed adapt based on circumstances.

Often, when there is room for a choice, the dierence between the alternatives will be very

small. The typical example is a hand with several ryanmen waits where we need to break one up.

It’s usually best not to hesitate and pick one. It might backre, but true skill nds its application

elsewhere than in this kind of choices.

10 CHAPTER 1. TILE EFFICIENCY

Example qwwwertsdzxcvb

Here, we want to make 123 sanshoku and break up either rt or vb. We might cut t,

draw e the next turn and get tilted. However, we should reevaluate the situation coolly and cut

w next, with the possibility of 234 sanshoku.

How to think about eective tiles

When deciding what tile to cut, we should consider what tiles are the eective tiles of each tile. A

tile is called eective when, if we would draw it, we’d keep it and discard something else. If we’d

discard it, it cannot be called eective, though it may appear so.

Example qweyuagghnm334

Here, the a can be used with s or d to create a group candidate. However, the resulting

shape would be worse than anything we already have, so the tile is useless.

There are several degrees in tile eectiveness (can make a joint, can make a group, can make a

joint into a better joint etc.).

However, among eective tiles, there is a clear dierence between the ones which move the

hand closer to tenpai and the rest. We say they reduce the shanten. In general, reducing the

shanten is greatly preferable. (The number of tiles that reduce a hand’s shanten is called the

tile acceptance. The tiles that do not reduce shanten but still improve the hand are called upgrades.

We will use these terms throughout.)

Example rtuuasdghjklvb→ cut r

When we draw the backre y, we can cut u and make a wider iishanten. (Although tile

acceptance-wise, there is no dierence with tvb) l is bad, because sanshoku is too far away

and we lose ittsuu.

Important factors in point expenditure

There’s a huge amount of factors one needs to consider when comparing possible discards (how easy

it is to complete groups and joints, how strong they will be when they complete (good wait, score),

defensive power etc.) In practice, considering the most important factors will lead to the correct

decision most of the time. We can’t simply make a list of all the factors and give them equal weight

or we might get overwhelmed by the unimportant factors and make unbalanced decisions.

For example, when thinking of how to build our hand, it generally doesn’t matter that

much what seat we are in. It’s true that as dealer we should emphasize speed, but this is

mostly because of renchan and tsumo payments.

This factor only becomes important when thinking about whether to push or to pull, because

the probability of getting a renchan or getting tsumo’d depends tremendously on the opponents.

Especially in the early game, we should rely more on other factors (the hand itself) in deciding

how to build our hand, with only a small bias toward speed. There is really no dierence between

North, West and South seats.

The old saying goes that the North seat shouldn’t call to not give the dealer more draws, but

it’s questionable how much one extra tsumo increases the expected point loss. (It also makes South

and West get less draws which further balances the eect.)

However, we should keep our own seat wind until last, and shift the haitei away from the dealer

when multiple opponents are in tenpai. The eect is small, but unlike withholding calls, comes at

no cost to ourselves.

Since round expected value 6= hanchan expected value, according to the point situation there

are situations to prioritize win rate (emphasizing defense with a hand that looks dicult to win)

and situations to prioritize score. However, in most situations (enough rounds left, no one close to

busting out), it’s not really necessary to be conscious of these, and we can ignore the scores and

play normally.

Iishanten peak theory

When thinking what the next draw will be, being likely to draw manzu because it’s cheap in

opponents’ discards is theoretically correct, but having drawn one, thinking that we’ll draw another

manzu because we have the “manzu momentum”, or on the contrary that we’ll draw a pinzu or

souzu to “even out” is a grave error. The same can be said not just of suits, but of numbers,

sequences and pairs.

The probability of advancing the shanten with a single tsumo p is

p =

tile acceptance tiles left in the wall

tiles left in the wall

with the average number of draws to advance being 1/p. Accordingly, the marginal value of

tile acceptance increases the lower tile acceptance is.

That’s why, from the point of view of isolated tiles, play to maximize the tile acceptance

when they form joints, and from the point of view of joints, play to make the bad shapes

complete more easily. In general, play to maximize future tile acceptance rather than

immediate tile acceptance, because there are less eective tiles closer to tenpai.

Fundamental rule of tile eciency

Play to maximize future tile acceptance closer to tenpai rather than immediate tile acceptance.

Example wti→ cut w or i

Cutting t gives the biggest tile acceptance right now, but it’s the easiest tile to make a good

shape with.

Example qqkllccv→ cut c

Example With >5 blocks, break the weakest block completely.

The advantage of winning rst

Mahjong is a game about scoring the most points. There are generally two strategies for this type of

games, namely scoring points and preventing opponents from scoring points. In mahjong,

TN: This one of the reasons why I don’t recommend the Euophrys eciency trainer.

12 CHAPTER 1. TILE EFFICIENCY

both are a factor, but the methods of preventing opponents form scoring points are limited. We

can not deal in, but this doesn’t prevent opponents from self-drawing or dealing into each other.

There is only one way to prevent those, namely by winning rst. In mahjong, scoring a win

directly prevents opponents from scoring and even lowers their score. Folding completely (betaori)

is generally not that good when we are close to winning ourselves. Therefore (and we draw a clear

line from what is commonly called tile eciency by the public), it is indispensable to learn tile

eciency to a high degree of precision.

1.1 Tile logic

Discard choice method

To win, we need four groups and one head (pair) (we do not consider chiitoitsu, kokushi musou,

nagashi mangan). To eciently create four groups and a head, it’s good to divide the hand into

groups and group candidates.

Classication of hand components

From most to least complete:

(1) Groups (面子, mentsu)

Sequences (順子, shuntsu): qwe

Sets (刻子, kōtsu): kkk5

(Kans) (槓子, kantsu)

(2) (Simple) joints

(搭子, tātsu): become a group in one move

Penchan (辺張): qw

Kanchan (嵌張): ad, sf become ryanmen with one tile type and are called outer

kanchan, dg, fhbecome ryanmen with two tile types and are called inner kanchan.

TN: I have chosen to use this term instead of “protoruns” like Daina Chiba for the following reasons:

• It’s much shorter (1 syllable versus 3).

• They consist of two or more tiles joined together.

• (Complex) joints can be used to form sets, not just runs/sequences as “protorun” would suggest.

• It ts into the metaphor of a mahjong hand as a body with four limbs and a head. A joint is necessary to

form a limb.

1.1. TILE LOGIC 13

Ryanmen (両面) xc

The ryanmen is commonly called good shape, the other two bad shape.

(3) Pairs (刻子, toitsu) rr11

One of these is necessary to form the head. With two and up, they can be considered joints for

sets.

(4) Isolated tiles (oating tiles) Single tiles which can become a joint in one move and a group

in two. Includes shapes like wert or rtty that combine a group with a oating tile,

which we’ll discuss later. In general, we don’t include these under the term group candidates.

(5) Complex joints A type of joints, consisting of a joint and a oating tile, such as qqw

(penchan pair), qee(kanchan pair), eer(ryanmen pair), etu(ryankanchan), eeerrt

(a type of ryanmen pair). We’ll discuss their many possible types in detail later.

(6) Complex groups (many-sided waits) These are actually a type of joints. We use this name

to refer to complex clusters of tiles where the division into groups, joints, pairs and single tiles is

ambiguous. Examples are dfghj or dddf. However, if there is already another pair,

the latter will usually be considered a set and a single f. We will analyze this shape later.

Block theory

Groups, group candidates and the head are generally referred to as blocks. In mahjong, we must

create 5 blocks. If we have 4 groups or group candidates, the candidates are sucient, if we have

3 or less, they are insucient. If more than 5 exist in the hand, we have too much (overrun).

There may be multiple ways to divide a hand into blocks. For example, we can analyze qe

tuo as qe + tuo or as qet + uo. The analysis of a tile cluster can depend on

the rest of the hand. For example ddgj is dd + gj or ddg + j or dgj + d

if we have another pair, but if dd is our only pair, only the rst analysis is valid.

However, reading xvbn as xv + bn instead of x + vbn, or zxcvb as

zx + cvb instead of zxc + vb, decomposing complete groups or expressly analyzing

the joint as weaker is not done.

Discard choice patterns

With the components of the hand falling into the framework described above, we can divide possible

moves as follows:

Unless counting fu.

14 CHAPTER 1. TILE EFFICIENCY

1. Cutting an isolated tile

2. Cutting 1 tile from a complex joint

3. Dropping a joint or pair

4. Dropping a group

We will usually have to decide between 1, 2 and 3. When deciding between moves of the same

class, we can construct an order of precedence of shapes and choose the best shape, unless there is

some connection to yaku or dora.

For comparison between dierent classes, or identically good shapes, or when several possible

ways to break up a complex shape exist, we’ll have to establish dierent criteria.

In this strategy guide, we hope to create an all-encompassing classication of comparison criteria

to enable us to nd the best move.

Isolated tiles and simple joints

In mahjong, we must discard the tile we think is least useful every turn. We will now investigate

the basic order of precedence of isolated tiles and joints. We will operate under the assumption

that we don’t want to call and don’t care about the hand value.

To reiterate: we dene as n-shanten

a hand which requires n draws to achieve tenpai. Including

chiitoitsu, n is always at most 6.

For a group shape, n is at most 8.

Since achieving tenpai is a

prerequisite for winning, and achieving 1-shanten is a prerequisite for tenpai and so on, we will as

a general rule play the move that reduces shanten.

There are in general 3 kinds of moves that reduce shanten (for a group hand):

1. Making a joint from an isolated tile

2. Making a group from a joint

3. Making the head from an isolated tile

On the contrary, there are 3 kinds of moves that increase shanten (shanten return):

1. Dropping a joint with insucient group candidates

2. Dropping a group (except for cutting w from qqwe in an otherwise headless hand)

3. Dropping the only pair

While 1 is sometimes ecient to replace a weak joint with a better one, 2 and 3 are generally

inecient and constitute a loss. Because a head is easier to create than a group, 3 is more common

than 2.

From the above, the combined order of precedence for hand composition is isolated tile <<

joint <<< only pair <<< group.

Note on inequality signs:

The values of n are pronounced ii, ryan, san, suu etc. In practice, shanten counts below 3 are rarely used.

Where n = 6 - #pairs

Where n = 8 - 2×#groups - (1 if there is a head) - min(4, #joints, not counting the head)

1.1. TILE LOGIC 15

<<< almost always worse than

<< generally worse than, with the converse being true in limited exceptions

< a dierence exists, but it’s subtle; there will be many cases

where the rest of the hand or the discard piles will lead to the converse

Comparison of isolated tiles

Guest wind < 1·9 << 2·8 << 3 to 7 <<< 3445 or 3456

Taking the aka dora into account, 3·7 < 4·5·6, since 4 and 6 can make an aka ryanmen, while

5 always makes an inner kanchan.

When considering calls and yaku, a yakuhai pair is better than ryanmen. Isolated terminals

can’t make a ryanmen, but 2 and 8 can, and more easily than a yakuhai pair can be made from

a single yakuhai. Therefore in general, 1·9 < isolated yakuhai < 2·8, but this depends on the

rest of the hand (with many bad shapes, the value of yakuhai rises).

Comparison of simple joints

Penchan << outer kanchan << inner kanchan <<< ryanmen.

Taking the aka dora into account, 35·57 < 46, 13·79 < 24·68. But waits closer to the outside

are easier to win on, so with no aka the ranking is reversed.

Taking the aka dora into account, 23·78 < 45·56 < 34·67. Without aka dora, 45·56 < 34·67

< 23·78.

Inner kanchan and ryanmen contain middle tiles and are clearly better than isolated middles. A

penchan is similarly clearly better than an isolated 2·8, but the comparison between penchan and

isolated middle tiles is not so clear. In general, 3 to 7 < penchan. (In theory, an isolated middle

tile will create a group faster than a penchan given at least 12 draws, but even in the very early

game we should prefer the low shanten number allowing us to pressure opponents. However, this

can change with discards or through the inuence of other shapes.)

Comparing pairs and bad shape joints

Since at least one pair is required to win, we only have to compare pairs with other components

when we have at least two.

Two pairs

The tile acceptance is 2 types, 4 tiles, so similar to a bad shape joint. Let’s call honors and terminals

A-class pairs, 2s and 8s B-class pairs, middle tiles C-class pairs. These classes have respectively 0,

1 and 2 upgrades to a ryanmen, so the total amount of upgrades is equal to the sum of the parts.

Furthermore, when such an upgrade occurs, the resulting shape is a ryanmen pair, which is stronger

than a simple ryanmen. There are also upgrades into kanchan or penchan pairs. With 1 pair, these

upgrades can’t occur.

Accordingly, the order of preference is penchan < 2×A < outer kanchan < A and B <

inner kanchan < 2×B = A and C < B and C < 2×C. Often, preserving the two pair

shape is good.

16 CHAPTER 1. TILE EFFICIENCY

Three pairs

When dropping one of the 3 pairs, the loss is only 1 type, 2 tiles, which is less than a penchan.

Even an isolated middle tile is better. But because it has more upgrades than a bad shape joint,

a 3rd pair is better than a bad shape with only 2 tiles left. When deciding which pair to break,

choose the one that has been discarded the most and is dicult to turn into a set. If there is no

dierence, retain the ones with the best upgrades. (The case where the pairs are part of a complex

shape will be discussed later.) Often, breaking up the three pair shape is good.

Four pairs

The tile acceptance is low, but because it’s 2-shanten for chiitoitsu, break up bad shape joints

rst, especially when the group shanten is high and there are many bad shapes. When going for

a group hand, the order of preference is similar to the 3 pair case. With 5 pairs, we usually go for

chiitoitsu.

Complex joints I

Basic complex joints

Penchan pair qqw qww

Kanchan pair ssf sff

Ryanmen pair ccv cvv

Ryankan dgj

Of these, only the ryanmen pair is called good shape, the rest are bad shapes.

Comparison of basic complex joints

All of them except ryankan include a pair, so if the hand doesn’t have another head, they don’t

work as joints.

However, even if we have no other head and a shape like wwe, we can easily get a head

elsewhere, so the e is much stronger than a regular isolated e. It has a ryanmen upgrade on

drawing a pair into any other bad shape joint, so it’s about as strong as a cvvbor cvbn

serial shape. However, a shape like wwr is not as good, since it only has extra upgrades into a

These techniques stand in clear contrast to occult players like Tsuchida Kōshō, who often make conscious plays

for chiitoitsu from 3 or even 2 pairs. This should under no circumstance be attempted by beginner and intermediate

players who have not yet had time to develop their sense of ow.

1.1. TILE LOGIC 17

bad shape joint. When we draw e, we can make the good shape wwer which can be seen as

either a group or a pair + joint, so it’s slightly better than a lone r.

A complex bad shape has a tile acceptance of 8 tiles (including another pair for the pair shapes),

which is the same as a simple ryanmen. However, if it remains at tenpai, it has to be broken down

into a simple bad shape, and is therefore weaker.

The order of precedence of complex joints of the same type is similar to that of simple joints,

with more central joints being better because they have more upgrades. When comparing wwr

with sff, the dierence is small and the rest of the hand plays a large role. When deciding

what to cut from wwrr, when we draw t, we get either wwrt or rrt. So when we

want to create just a group, we cut w, but when we want a group and a pair, we cut r.

When comparing complex joints, we don’t want to get rid of them entirely like when comparing

simple joints or isolated tiles, but drop one tile and turn them into a simple joint. To get the

optimal wait in the end, drop one tile from the strongest shape as a general rule.

However, when breaking up a complex joint that contains a pair, there are situations where we

want to x into a simple joint, and situations where we want to x a pair. We will discuss the

dierence below.

With two pairs

Fix the bad shape into the head. With two ryanmen pairs, turn one of them into the head,

depending on dora, yaku, discards etc.

With three pairs

One complex joint Drop the pair from the complex joint.

Two complex joints If we have a ryanmen pair, x the ryanmen. If both the complex

joints are bad, x the one with the most ryanmen upgrades as a general rule. However,

in qqwhhm.., the simple pair has good upgrades, while the complex joints do not,

so temporarily cut w, hoping to draw gj. This is even more applicable if the simple pair is

connected like dfghh.

Three complex joints If all the complex shapes are bad, x the worst one into the head.

This gives a momentary loss of 2 tiles of tile acceptance, but creates a better shape later on once

one of the joints completes into a group.

With one ryanmen pair, x the ryanmen. With two ryanmen pairs, x the bad shape into

the head. With three ryanmen pairs, there is no big dierence, so consider the usual suspects of

discards and dora.

With ryankan and ryanmen pairs, x the ryanmen. With two ryankan, try to x one towards

accepting an aka dora, cutting w from wry, leaving the upgrade into ryanmen with u.

Otherwise, try to create a suji trap by cutting g from adg.

18 CHAPTER 1. TILE EFFICIENCY

The comparison between ryankan and paired bad shapes is dicult. With a three pair shape,

x the strongest kanchan, hoping to draw a ryanmen upgrade. If already iishanten, the upgrade

is less important and we can x the ryankan into a desirable kanchan as above.

Complex joints II

When comparing complex joints, the presence of connected groups or pairs can create extensions.

An extension is stronger than the original shape, but cutting a tile from the extension leads to

a bigger loss. We therefore want to keep extensions alive as much as possible, an exception to

the principle of xing strong shapes. There are many kinds of complex joint extensions, with the

following being the chiefest.

Penchan pair extensions

aassdf aaassd

Kanchan pair extensions

ssfghj sffghj dghjkk

adddfg dddfgg

Ryanmen pair extensions

sdfggh ddfggh dfgghh

dfgggh fggghh

Ryankan extensions

sfghjl(long ryankan)

sfghhj(ryanmen kanchan)

ssfhh(ryankan pair)

aadfghkk(long ryankan pair)

sddfghjjk(double ryanmen kanchan)

These are really important and you should denitely memorize these, especially the rst three.

1.1. TILE LOGIC 19

Using complex joint extensions

Like the ryanmen kanchan, complex joints can be dicult to notice if you’re not used to them,

so it’s important to pay close attention. The ryanmen kanchan has the same tile acceptance as a

ryanmen pair, but is guaranteed to produce a sequence, so we should prefer it over a ryanmen pair

in a pinfu hand.

Bad shape extensions tend to have many upgrades into a good shape. We therefore want to

preserve them. In a two pair shape, we x the regular complex joint into the head, keeping the

extended shape. In a three pair shape, we drop the pair from the extension.

With complex joints that contain 3 identical tiles, the ryankan pair and the double ryanmen

kanchan, the loss from breaking them up is especially big, so we prefer to keep them. With multiple

such shapes, we drop 1 tile from the one most likely to give a good shape.

Example ertttyaadggm,. → cut t

In a three pair shape, the choice between xing ertttyinto a good shape and dropping

a tile from a complex bad shape is dicult. If the bad shape is an inner kanchan, we x it and

hope for ryanmen upgrades. Otherwise, we x the ryanmen.

Example ertttyaadllbmm→ cut m

Example ertttyaadllm.. → cut t

Adding to the loss that comes with breaking up extensions, they can easily progress into better

extended shapes by drawing more tiles.

Example eerrtyyujklmm,→ cut m

However, memorizing all these dicult shapes takes a lot of time and isn’t really necessary. If

we remember that joints become stronger when they’re connected to groups or pairs, we can deal

with most hands correctly.

If possible, we should also try to remember the shapes that lead to the strongest extensions by

cutting one tile (aaddgg cut d) or by drawing one tile (adgg draw a). Especially

the latter are important because they are easy to overlook.

As a general rule, to create strong shapes, we also want to keep their precursors, and the

precursors of those precursors. This will later allow us to classify the relative strength of simple

joints and isolated tiles depending on their surroundings.

Ultra bad shape joints and furiten joints

So far we have not considered what tiles have been discarded yet. We will now discuss the most

important inuences of discarded tiles on tile eciency, namely depleted tiles and furiten.

Ultra bad joints (bad shape joints with 2 tiles or less left)

Since the marginal value of tile acceptance increases when it’s low, the dierence between 4 tiles

and 2 tiles is rather big. When in tenpai, it’s quite painful to have an ultra bad shape left (a head

start riichi in the mid game with such a wait will deal in more often than win). We prefer a 3rd

pair over such a joint. We also prefer an isolated middle tile (or even a 2 or 8 if the depleted joint

is a penchan), breaking the depleted joint and trying to build another.

20 CHAPTER 1. TILE EFFICIENCY

However, a bad shape joint with 3 tiles left is not that bad. Also, we prefer a ryanmen with 4

tiles left over a bad shape with the same amount of tiles left, since we know opponents don’t want

to use those tiles.

An extra pair of which the other two tiles have been discarded can be treated as a oating tile,

so it’s also not as bad as a depleted joint, especially if it’s a middle tile.

Unless 3 or 4 tiles of the same type have been cut, discards don’t really matter for comparing

isolated tiles, since the dierence in ease of creating a good shape between a 1, 2 and 3 is more

important.

We should also consider upgrades of bad shapes that have become depleted. For example, if h

has been cut 3 times, dg becomes dicult to turn into a ryanmen, and a ryankan with j is

also disappointing.

All these make it that remembering to look at the discards has a rather big impact on results.

Furiten joints

We can’t ron while in furiten. But this only matters in tenpai, and a furiten good shape has a win

rate not much lower than a regular bad shape. Accordingly, we should prefer furiten ryanmen over

bad shapes. Furiten bad shapes are bad, and we’d rather have an isolated tile. (Unless it’s already

the end game and we want to collect no-ten payments.) Similarly, an isolated tile that can make

a furiten joint is worse than one of the same class, but better than one of a lower class. When

choosing how to break up a complex joint for riichi, we usually pick the wait that isn’t furiten, even

if it’s narrower.

For example, cutting q from qw and drawing e, we should keep it and go break up a bad

shape (that’s why we usually break bad joints from the outside). Cutting q from qa and

drawing w or e, we keep it and cut a.

Comparing components of the same class

The interaction of tiles

Apart from neighboring tiles being heavily discarded or being furiten, isolated tiles and joints of the

same class can still be better of worse through the presence of other components (especially joints)

that have an overlap with them. This means that their tile acceptances or upgrades compete for

the same tiles, weakening them, or that they will compete for tiles once an upgrade occurs. For

example, in wt, the two isolated tiles compete for e and r.

We can also have combos, which are shapes that have the same tile acceptance as a regular joint

or isolated tile, but are easier to upgrade or have upgrades to a superior shape. For example, the

m in cvbm is stronger than a regular isolated m, because when we draw n, we get the

strong cvbnm shape.

There are also cases where an overlap and a combo exist at the same time. In general, if there’s

an overlap, the shape is weaker than usual even if a combo exists. A typical example is

a eryu shape which has an overlap on the t, but creates a strong combo once we draw it.

1.1. TILE LOGIC 21

An exception to this general rule are shapes where the overlap is for a bad shape-creating tile and

the resulting combo is a better shape than normal: for example, z and b both become stronger

than usual in a zb shape, since we can draw c for ryankan.

When comparing overlaps and combos, their quantity and quality are both important, but in

the case of an overlap and combo existing at the same time and weakening the shape, quantity >

quality since we want to draw the combo as soon as possible.

In general, we also prioritize overlaps and combos that exist right now, rather than those that

will be created when an upgrade occurs.

Diagram of component precedence

We can roughly summarize preference between components (joints and isolated tiles) of the same

type as follows:

Compare overlaps in the tile acceptance

↓

Compare the strength of the shapes that are created when a tile is accepted

↓

Compare overlaps in the upgrades

↓

Compare combos with other components

Like this, we get the following order: overlap in tile acceptance < overlap in tile ac-

ceptance, but once the tile is accepted, the shape is stronger than normal < weak

overlap in tile acceptance < overlap in upgrades < overlap after an upgrade < normal

< combo after an upgrade < weak combo < strong combo.

A weak component of a class can be compared with a component of a lower class (weak 3 versus

2), a strong component with one of a higher class (strong outer kanchan versus inner kanchan).

By understanding what joints are weak and strong, we can understand what isolated tiles are

good or bad at making strong joints. We’ll therefore rst compare joints, and continue with isolated

tiles.

In the following sections, there are some points of attention to keep in mind.

• We’ll assume that comparison of joints to joints takes place in an overrun situation (>4 joints),

that of isolated tiles to joints in a non-overrun situation. We’ll come back to overruns later.

• When not stated otherwise, assume we have a head elsewhere in the hand.

• There will be some shapes that give rise to iipeikou, put we will consider only eciency and

disregard the possible extra score.

As always, there are exceptions when overlaps and combos are dicult to analyze, but the above

should be a good guideline. Now, because this theory can be quite perplexing when put into words

abstractly, we’ll put in it into practice on the most common shapes.

22 CHAPTER 1. TILE EFFICIENCY

Penchan

Summary

qwwr< qwry< qwrr< qwrt< e< qwtt< qwty

< qwyy< qwyu< qw< qwtyu< qwrty< qqwwe

Tile acceptance overlap

qwwr

To make two groups here, we need to draw e twice, and we will commonly cut q to create a

wwr kanchan pair. The q is essentially useless, and we’ll cut it over an isolated a.

qwry

It’s easier than the above shape to make two groups, but still not desirable. By cutting q, we get

the ryankan wry. The q is not worth keeping over an isolated s, or even a unless the rest

of the hand is very bad.

qwrr

Similar to the above shape, but if we draw the e, the resulting qwerr is a bit stronger

than qwery, but because of the overlap (we can cut q, drawe and still end up with a

group + oating r), this shape is still weaker than just qw.

qwrt

Drawing e instantly, we get qwert group + ryanmen, which is a good shape. All of the

above penchan are inferior to an isolated d. In general, any bad shape joint with a tile acceptance

overlap is worse than a oating middle tile.

qwtt

No direct overlap, but if we drop the penchan and draw the e backre, we still get the kanchan

pair ett. The pair can also upgrade into the above shape.

Upgrade overlap

qwty

Only overlaps for upgrading the penchan into wr kanchan. Not a big dierence from a regular

1.1. TILE LOGIC 23

penchan.

qwyy

Compete for r to upgrade into wr and ryy.

Post-upgrade overlap

qwyu

When we draw r and upgrade, there is an overlap for the next upgrade with t.

We might even say that a qwui shape is worse than a normal qw, because we draw r,

then t and get overlapping ryanmen rtui... But that will almost never happen in practice.

In general, when tiles are separated by at least 5, we can safely ignore their inuence on each other.

Normal

Upgrade combo

qwtyu

Drawing r, the wrtyu shape has many upgrades into ryanmen.

qwrty

Transforms into the above shape on drawing u.

However, because a penchan can’t upgrade into a ryanmen directly, at 1-shanten, we should cut

a from qwasfgh, since we can draw the backre d and make a furiten sanmenchan.

This is a penchan-only exception, and from wrsfghj we cut w.

Combo

qqwwe

Makes a complex joint when drawing any of qwr. (The qwwer shape can be considered

as qwe + wr and will be discussed under outer kanchan.)

We will often want to drop a penchan to keep a oating tile. The following shapes make this

especially eective:

24 CHAPTER 1. TILE EFFICIENCY

tio

Overlap for yto upgrade the penchan/complete a ryanmen, and for uto complete the group/com-

plete a kanchan.

rio

Overlap for y to make ryankan.

Outer kanchan

Summary

adfh< adff< aasf< addg< addf< adgh<

adhh< adhj< ad < adhjk< adgg(g is the only head)

< aasdd< addfg< assdf< ssdff< adghj

< adfgh

Tile acceptance overlap

adfh

The hardest shape to make two groups in, we often want to slim this down to just df and make

one group. We’ll still keep the a much more often than the q in qwwr.

adff

Similar to the above, but becomes a head + ryanmen pair if we draw another a.

aasf

Unlike the above, we can upgrade into a ryanmen by drawing g. In real matches, dropping the

aa and going for tanyao is common.

addg

Cut a for kanchan pair or d for ryankan, a similar idea to q from qwry.

addf

Instantly becomes good if we draw s. With all of the above, we’ll often cut one tile and prefer a

1.1. TILE LOGIC 25

oating middle tile, similar to penchan.

Upgrade overlap

adgh

Overlap for ryanmen upgrade and group completion.

Post-upgrade overlap

adhh

Kanchan pair with f.

adhj

Overlapping ryanmen with f.

Normal

Upgrade combo

adhjk

Easy to upgrade ryankan with g.

Combo

adgg(g is the only pair)

Ryankan pair with a. If we have another head, worse than a regular zc, cutting a or g.

aasdd

Strong kanchan pair with ad, ryanmen pair with f.

addfg

Upgrades into a complex joint with any of adgh.

26 CHAPTER 1. TILE EFFICIENCY

assdf(penkanchan)

Upgrades into a complex joint with any of asfgh.

ssdff

Upgrades into a complex joint with any of asfgh, with better shape post-upgrade than the

last one.

adghj

Sanmenchan with f, ryankan not only with g but also with k. Still not as good as an inner

kanchan, which has 2 ryanmen upgrades.

adfgh

Upgrades with f for ryanmen, g for ryanmen kanchan, j for sanmenchan, k for long ryankan.

Clearly better than an inner kanchan.

Inner kanchan

Summary

cbnn < cbbm < cbbn < cbm, < cb < zxcvn <

xcvvn< xxcb(x is the only head) < ccvbb < cbnm,

Since an inner kanchan is just an outer kanchan with one more ryanmen upgrade, the explanation

has been omitted where similar to the previous section.

Tile acceptance overlap

cbnn cbbm cbbn

The dierence from the corresponding shapes with an outer kanchan is that these can become

ryanmen, so it’s common to keep these over a oating tile. Because they reduce tile acceptance, we

will often drop these for a penchan or outer kanchan. (But when a good end shape is important,

we’ll keep these.)

Upgrade overlap

cbm,

Normal

1.1. TILE LOGIC 27

Combo

zxcvn

No extra ryanmen upgrades, but easier to make a paired joint.

xcvvn

Can draw v for good shape, so slightly better.

xxcb(x is the only head)

Ryanmen upgrade not just on xn, but also on b. Similarly, a xxcbm ryankan has an

extra ryanmen upgrade on m. In general, tiles next to the only head become better once a head

completes elsewhere in the hand.

ccvbb

Becomes a strong complex joint with c or b.

cbnm,

Very easy to upgrade into ryanmen.

Ryanmen

Summary

wwrtuui(with no other head) < errtyyu< rrtt< eryu

< eety(with no other head) < er< qwwee< qweer< wweerr

< qwert< weert < werty

Because a ryanmen is so strong, the advantage or disadvantage of overlaps and combos is rather

small. Accordingly, when comparing ryanmen, we should bear in mind that the importance of

discards relative to the hand shape increases.

Tile acceptance overlap

wwrtuui(with no other head)

An extremely unusual shape, since cutting t causes no loss in tile acceptance.

errtyyu

By cutting r or y, this overlapping ryanmen shape becomes a sanmenchan.

28 CHAPTER 1. TILE EFFICIENCY

rrtt(double ryanmen)

Double overlap on ey. Cutting one tile gives a ryanmen pair.

Even with a bad overlap like this, the nal wait when in tenpai will be good, so we prefer this

kind of shapes over simple bad shape joints. Still, we prefer a ryankan over this shape.

eryu

Overlap on t. Drop this over another ryanmen.

eety(no other head)

Cutting y is still salvageable when we draw r.

Normal

Combo

qwwee

Drawing w or e, we get a complex joint that waits on qwer, but in practice it’s harder to

draw on vertically than a normal ryanmen.

The following are variations of the same shape that are somewhat easier to draw:

qweer

wweer

qwert

weert

Can make a ryanmen kanchan with u.

werty(sanmenchan)

The strongest possible 5-tile shape.

1.1. TILE LOGIC 29

Pair dropping

We have touched on this topic in the section on comparing pairs and bad shape joints, but we can

also compare pairs to each other.

Pair dropping process

We usually think of dropping a pair when we have at least 2, and we have seen earlier that in a

hand with exactly 2, dropping a bad shape joint instead is common.

However, in a hand with few groups and ryanmen and many bad shape joints, there are cases

when dropping one of the 2 pairs is preferable over dropping a joint. The reason is that in a

high-shanten 2-pair hand, the probability of drawing a 3rd pair before tenpai is high, creating

an undesirable 3-pair shape, which we would try to break by cutting a pair from a paired joint.

Therefore, instead of waiting for that to happen, we drop a terminal pair in advance, hoping to

create a superior 2-pair shape later on, following the theory of emphasizing tile acceptance closer

to tenpai over immediate tile acceptance.

Example weyiaaf;kkxvn,dora k

→ instead of slimming down souzu, drop a and start forcing tanyao

With 3 pairs, we usually want to break a pair and return to a 2 pair shape. However, if we’re

certain the future 2 pair shape will be suciently good and don’t need any further improvements,

we can preserve the 3rd pair to maximize tile acceptance right now.

Example wweuddfhjkcv22→ condently cut u

In 3-pair hands, if we have ying pairs like wwrr or kkll, we should preferentially

cut 1 tile from these, to preserve the horizontal tile acceptance. However, if we have at least 4 pairs,

we should preserve this shape since it’s a many-sided wait and can gives us iipeikou. Whether to

cut the inside tile or the outside tile from this kind of shapes has been discussed earlier: if we are

content with a group, cut the outside; if we need a group + a pair, cut the inside.

If we don’t have any ying pairs, we should break the pair that has the weakest upgrades into a

joint, so usually the most outside pair, comparing the pairs the same way we compare isolated tiles.

However, from qqttaa, we should denitely cut a, since we can draw e for a ryankan

pair. We will discuss the relative worth of isolated tiles next.

Ukase-uchi

Up until now, we have discussed dropping the most unneeded pair from the hand entirely. There’s

another kind of pair dropping, where we cut only one tile from a (preferably middle) pair, down-

grading it into a oating tile and hoping to stick tiles to it horizontally. This is a special move

called ukase-uchi. We use it when we want to upgrade to a sticky shape, or when the remaining

joints are all too weak to make a decent hand. We can also use it when we have sucient ryanmen

joints, to create yaku, or in shapes like qquommm,. (1 q left) where a shanpon wait

would be extremely painful.

These examples were not present in the original, I added them to illustrate the idea more clearly.

30 CHAPTER 1. TILE EFFICIENCY

Isolated 1 and 9

We can compare isolated tiles using the same method we used to compare joints, considering overlaps

and combos. However, because there many more possibilities to draw up to a group, there will often

be many possible overlaps and combos in dierent future shapes, making comparison more dicult.

It will be good to consider these sections more as a general outline that can be complemented by

information from discards and desired yaku.

Summary

qru < qrr < qr < qry < isolated guest wind < qrt < qwr

< qwwr< qtu< qty < q< qer< qqwe< qtyu <

qrty< qrtty< qt< qery< w< qeryi< qert

Tile acceptance overlap

qru

Even after we draw ty for qrtyu, there’s still overlap between q and r, so q is

unnecessary.

qrr

Can make a complex joint wrr or err, so weaker than just qr. Even if rr is the

only head and we can try to make a group with the q, it’s still clearly weaker than a normal q.

The famous shape with an overlap for we. Even though we can draw towards qwer, a

good wait in a headless shape, it’s still clearly weak.

qry

Ryankan on w, ryanmen + oating tile on e. The loss from missing out on qery doesn’t

really hurt.

With all of the above, we prefer to keep an isolated guest wind because of its defensive power

(especially if we can draw a pair).

qrt

Drawing w or e creates an overlapping shape. But the loss from missing out on a qert is

1.1. TILE LOGIC 31

quite big.

qwr

Overlap on both w and e, but can draw the qwer serial shape, which is better than the

above, especially when we have no head.

qwwr

As discussed earlier, it’s usual to see this as q+ wwr. But we can miss out on the qwwer

shape which is slightly better than the above.

qtu

Only overlaps on e for ryankan, which makes it better than all of the above.

Post-upgrade overlap

qty

After drawing e, qety has an overlap for r.

Normal

Combo

qer

Has many possible overlaps and combos, so dicult to compare. When there is no head, the

qwer shape is valuable, so in that case it’s somewhat better than a lone q.

qqwe(aryanmen)

Group + head with q or r, connected shape with w or e.

qtyu

The kanchan with e has better upgrades than normal.

Do not confuse with the Aryan Men local yakuman (qriiafkkzv,,55)

32 CHAPTER 1. TILE EFFICIENCY

qrty

As above but better.

qrtty

Can draw e for ryanmen kanchan.

From here on, if aiming for a closed hand, it’s common to to keep these over a single yakuhai.

Makes a ryankan with e. In the above shape, cutting q and drawing e still gives a ryanmen,

while here we get a simple kanchan which is a bigger loss.

qery

It’s usual to see this as q + er + y, but drawing t creates a strong qerty kanchan.

From here on, the value is higher than an isolated 2.

qeryi

Can make a ryankan (quasi-good shape) with w or t, while a lone s can only make a ryanmen

with d.

qert

Ryanmen with w, strong kanchan with y, clearly better than a lone s.

(we’ll analyze the qwer shape as qwe + r in a later section.)

Isolated 2 and 8

Summary

sfg < sgj < sfgj < sgg < sgk < sg < shk <

sgjl< shj< s< sfgjl< shjk< sghj< sghhj

< sfgh< ssdf < sh< d

1.1. TILE LOGIC 33

Tile acceptance overlap

sfg

Overlap on d, but we can miss out on sdfg.

sgj

Overlap on d and f, but we can miss out on the sd ryanmen, so stronger than the above.

sfgj

Similar to qery.

sgg

Similar to sg, but weakens the s since the kanchan pair dgg is stronger.

sgk

Same logic as qru<qr. But because it can make a ryanmen, it’s better than the q in

qt.

Overlap on d and f.

All of the above are usually considered weaker than an isolated yakuhai.

shk

Overlap on f, but not on d.

sgjl

Appears to have an overlap, but the sfgjl shape is not that bad.

Post-upgrade overlap

shj

After drawing f, has an overlap on g.

34 CHAPTER 1. TILE EFFICIENCY

Normal

Combo

sfgjl

Same as qeryi.

shjk

Can draw f for sfhjk, which gives a sanmenchan with g and ryankan with l.

sghj

Strengthens the kanchan with f.

sghhj

Ryanmen kanchan with f.

sfgh

Sanmenchan with d, strong kanchan with j. Still not as good as a middle tile with 2 upgrades

into ryanmen.

ssdf

Easy to connect shape. Much stronger than a qqwe shape. While sfgh makes it easy

to create a kanchan, this one makes it easy to create a pair. Depending on what we want (how

much pairs we already have), either can be better.

Ryankan with f. Still not as good as a middle tile.

Isolated 3 to 7

Summary

xcbm,< x< cbn< cn,< ccn< zzv< cn< cmm

1.1. TILE LOGIC 35

< cm. < cm,< c< zxcc< cnm,< cnmm,< zxcv

< cm< xcvv< cbnm< ccvb< penchan < cvvb< xcvb

< cvbn< ccvbnn(no other head) < zxcvbnm< xcvbnm,

Tile acceptance overlap

xcbm,

Has an overlap on both sides. Pretty much the only case where a middle tile is worse than a 2.

cbn

Overlap on the v, but misses out on the cvbn shape.

cn,

Overlap on v for ryankan and b for ryanmen upgrade.

ccn

Overlap on the v and b for a complex joint.

zzv

As above, but the possible complex joints are not as good.

Overlap for simple joints, so not as bad as ccn.

cmm

Overlap on b for kanchan pair.

cm.

Overlap on b for ryankan.

Post-upgrade overlap

cm,

Similar to qty.

36 CHAPTER 1. TILE EFFICIENCY

Normal

Combo

zxcc

Easy to make a serial shape.

cnm,

Strong kanchan with b.

cnmm,

Ryanmen kanchan with b.

zxcv

Serial shape.

Ryankan with b.

xcvv

Easy to make a serial shape.

cbnm

Sanmenchan with v, strong kanchan with ,.

ccvb

Easy to make a serial shape, more likely to give good shape than xcvv.

The patterns from here on make it extremely easy to create good shapes, and are much

stronger than any other oating tiles. They are often better than a penchan, and comparable

to an outer kanchan.

1.1. TILE LOGIC 37

cvvb(nakabukure)

Ryanmen with any of xcbn.

xcvb(connect 4 shape)

Ryanmen with zcv, sanmenchan with n, strong kanchan with ,.

cvbn(connect 4 shape)

Even better than xcvb, makes sanmenchan on xm.

ccvbnn(no other head)

Makes a ryanmen with any of xvbm and is easy to extend further.

zxcvbnm(connect 7 shape)

Ryanmen with vb, sanmenchan with c, and can give ittsuu. We will often cut a tile from

ryankan or a kanchan pair to try and extend this shape.

xcvbnm,(connect 7 shape)

Ryanmen with cm, sanmenchan with zvn. and can give ittsuu. The best possible oating

tile.

Joint dropping

So far, we have discussed that there are four possible discard choices in a mahjong hand:

1. Cutting an isolated tile

2. Cutting 1 tile from a complex joint

3. Dropping a joint or pair

4. Dropping a group

Of these, dropping a group is a special move that only makes sense when chasing value, so we

will not discuss it here. We have already considered comparisons between moves of the same type,

as well as (in part) comparisons between 2 and 3. From now on, we will discuss comparisons be

that fall into the remaining cases.

When deciding to drop a simple joint, we do not only have to decide which joint to drop, but

which of its tiles to cut rst. We will now examine this question.

38 CHAPTER 1. TILE EFFICIENCY

Joint dropping process

It’s often easy to decide what tile to cut rst when dropping a joint. When we don’t have enough

group candidates, we want to keep the tile that has a chance of making a good joint while isolated.

That’s why we usually cut joints from the outside in. When we have sucient group candidates,

we can also consider the added value from yaku or dora that the tile we keep allows us to create.

Finally, there’s the defensive value of the remaining tile. In general, outside of mid -late games

where the end is in sight, upgrades are more important than defensive power.

There are some dicult situations where cutting a tile while leave a complex joint, and there

are several dierent complex joints we can choose to retain.

The example of wwrr has already been discussed.

From eetu, we can make a ryankan or kanchan pair. If completing the ryankan would

give us pinfu, we should prefer it. Otherwise, we’ll choose the shape with the most upgrades into

ryanmen. If we are able to call, we should keep the kanchan pair to call pon.

Even in pinfu shapes like qweuiddfghccbm, we should take the kanchan

pair if the head has many upgrades into a ryanmen.

Problem wrrtyyu→ cut w or r?

In a pinfu shape, cut r, else cut w to make iipeikou. When calling, cut w. (With tanyao

pinfu cutting wis not bad.) From wrtyyyu and wrtyyuu, we clearly cut w.

qetu wryi(sankanchan)

qetuo(suukanchan)

These shapes are not that dicult, but can turn into more complex shapes like qeetu

(sankanchan pair).

Though we can analyze them as complex joints, as a whole they form exceptional shapes that

we have not analyzed yet. (Cutting a tile from in a non-overrun increases the shanten.) Let’s try

to compare their strength to other types of components.

A sankanchan has no overlaps, but if we cut one tile from it we end up with a ryankan, so

it’s easier to break up than an outer kanchan. A qetu shape can upgrade into a ryanmen

with i, and we prefer it over an isolated middle tile, but cutting a tile from wryi to keep a

oating tile can be done. A suukanchan is stronger than a sankanchan, about on the same level as

an outer kanchan, but worse than an inner kanchan.

1.1. TILE LOGIC 39

Sankanchan and suukanchan pairs

The author considers this not as important and refers to the following blog post (日本語): https:

//chirno.hatenadiary.org/entries/2008/05/18

Comparing simple and complex joints

The general rule of tile eciency is to maximize tile acceptance close to tenpai. Often,

cutting a tile from a complex joint will maximize the immediate tile acceptance, but narrow the

shape down the road. Accordingly, in an overrun (>4 joints), we should break a simple joint

completely. Obviously, we should pick the weakest one. We should not ignore the possible benets

of leaving one tile (the surplus tile) when deciding which of the tiles to cut rst.

There are the following exceptions:

The complex joint is lower class than the weakest simple joint

For example, when we have the simple joints are all ryanmen and we have a kanchan pair. Since

we want to keep the ryanmen, we should cut a tile from the complex joint, in such a way that if

it completes, we get the most desirable result. (For example, cutting o from ioo.) But in a

hand like qqettdfxccvbm,, we cut ,.

The shape we get when the weakest joint completes is very strong

Because it gives us yaku or dora. With yaku, we can often cover the loss in speed by calling.

Example ioghjklzzxm,55→ cut x

In an overrun, we will often compare simple joints with oating tiles. Since dropping a joint

doesn’t reduce shanten, we will often want to keep a good oating tile instead of a joint. (Whether

to keep a safe tile over a joint is dicult.)

For example, we’ll keep a oating middle tile over a penchan, and a serial shape over an inner

kanchan. We will sometimes even drop ryanmen to keep dora or create yaku. In a headless overrun,

we’ll often drop bad shape joints over aryanmen like uioo and not care too much about the

head.

Example qeradghcvbb111dora q → cut a

Example weuioosdjlxcvb→ cut l

At iishanten, whether to take the straightforward shape or to drop a joint and try to get a better

wait from a oating tile is dicult.

In general, these are slightly better than a regular shape but not by much. For example zccbm33

accepts 1 more tile type than zccgj33, and zccbmm. accepts another 1 type more.

This is an exception to the exception, I don’t know why the author chose to include it. It’s the correct move

because the ryankan pair on manzu is strong, we can cover n with a ryanmen kanchan, and we preserve 345

sanshoku.

40 CHAPTER 1. TILE EFFICIENCY

Example qetyaaszcbnm222dora b → cut sor qw

Example qqetyjkxvnmm,. dora r→ cut e

We have enough ryanmen, so the advantage of drawing w is small.

In an overrun, we prefer to cut 1 tile from a joint in such a way that the surplus tile forms a

complex joint, like from qeet. But we still want to emphasize tile acceptance close to tenpai:

Example tysdghzccb,.44→ drop ,.

Comparing isolated tiles and complex joints

Cutting a tile from a complex joint and keeping a oating tiles with insucient group candidates

makes the joint a bit more dicult to complete, but makes it possible to create another group

candidate with the oating tile, increasing the win rate. It is nearly always limited to xing a

ryanmen pair into a ryanmen, or a bad shape paired joint into the head (other cases are more often

than not inecient).

Since it’s usually more dicult to make a group from a joint than joint from an isolated tile, we

only do this when the loss from the oating tile is very big. Also, a ryanmen pair can often end up

as two groups. (Especially drawing t into err is very big.) With wwe, it’s easier to form

two groups than with an isolated w, but an isolated e is still preferable. With eer and up,

the loss in eciency is big, and it’s good to keep the extra tile over isolated tiles. In shapes like

vv/ or xccvbn, the gain from completing a set is smaller and we will often slim down

these shapes to keep oating tiles.

Example wweoocmm,.gghj→ cut w

There are also situations to x the head. When xing a ryanmen turns a bad shape paired

joint into a head + oating tile, we can consider if it isn’t easier to create another group elsewhere

(wwejll → cut w, errkll → cut k).

When we have a oating tile in a nakabukure or connect 4 shape, we’ll often x a ryanmen to

preserve it. Still, we’ll often preserve a bad shape complex joint. We’ll only cut from one when it

moves us towards yaku (uoosdfgkkcvnm, → cut o) or to preserve a connect

7 shape.

These decisions will change a lot depending on yaku and dora, so we’ll want to give special

thought to those in later chapters.

Complex groups (many-sided waits)

Using complex groups

A shape like errr can be used a 3-sided wait if it contains the head, but if we already have

a head, it’s more likely to be seen as rrr and a weakened e that is more dicult to make a

sequence with.

1.1. TILE LOGIC 41

With this shape, we can have the decision at iishanten whether to take the irregular wait or

to take a very wide headless iishanten. We will return to this topic in the chapter on iishanten

eciency.

It’s not really necessary to think about the relative desirability of complex groups, but it can

be dicult to see what the actual wait is. Dicult waits come up rarely, but being able to solve

them quickly is directly tied to good results.

Classication of waits

1 tile 3 waiting on 3tanki

4 tiles

With 0 sets

qw33waiting on epenchan

wr33waiting on ekanchan

ww33waiting on w3shanpon

wertwaiting on wtnobetan

erttwaiting on wtaryanmen

we33waiting on qrryanmen

With 1 set

qqqwwaiting on wepenchan tanki

qwwwwaiting on qepenchan tanki

etttwaiting on erkanchan tanki

rtttwaiting on eryryanmen tanki

7 tiles

With 0 sets

When not playing with tips like a middle schooler :^)

42 CHAPTER 1. TILE EFFICIENCY

qwertyuwaiting on qrusanmentan

qwerrtywaiting on qru

werty33waiting on qrusanmenchan/piano wait

With 1 set

weeertywaiting on qwru

eeerrtywaiting on wrtu

eeerttywaiting on rtu

eeertyuwaiting on wrtuithe famous 5-sided wait

eeertyiwaiting on ui

eeettyuwaiting on rti

eeetyuiwaiting on rti

eerrrttwaiting on ert

eeerrttwaiting on erty

eeert33waiting on ey3entotsu/chimney wait

With 2 sets

wwwerrrwaiting on qwerttatsumaki/sandwich wait

wwweeerwaiting on wertopen sandwich wait

wwwryyywaiting on ert

Using 4 tiles

eerttttwaiting on ery

wweeeerwaiting on qwrt

1.1. TILE LOGIC 43

weeeertwaiting on qwrt

10 tiles

eerrttyyuuwaiting on eryu

eeertyuuiiwaiting on yuio

eeertyyuiiwaiting on tui

eeertttyuuwaiting on tyu

eeerrrttuuwaiting on wtu

eerrrttyuuwaiting on wty

eeerrtttyywaiting on rtyu

eeertyuiiiwaiting on wertyuio

eeerttyuuuwaiting on ertyu

eeertyuuuiwaiting on eyuio

And as a bonus, here are some 13 tile shapes that wait on any tile in a suit:

weeeertyuuuui

werrrrtyyyyui

wwwertyyuuuui

qqqwertyyyyui

wwwertyuuuuio

qqqwertyuiooo

Solving many-sided waits

Tha author refers to very a long blog post by someone else at http://ameblo.jp/010101/entry-10514689527.

html. The best way is of course to practice, for example with https://www.gamedesign.jp/flash/

bamboo/bamboo.html, the chinitsu problem book by Baba Hirokazu or actual tiles.

In the following chapters, we will discuss

• Eciency taking into account score and calls

44 CHAPTER 1. TILE EFFICIENCY

• Riichi and calling judgment, to understand the balance between score and speed

• Details of yaku and dora usage, ditto

• What tiles to keep for calling and yaku, how to choose discards with the above knowledge

• Iishanten eciency

1.2 Head start tenpai judgment

Understanding tenpai eciency

This isn’t really tile eciency, but we’ll take a look at whether to take head start tenpai. (Head

start means we are the rst player to tenpai.)

The reasons for including this in the tile eciency section:

• By understanding the strength of head start tenpai (riichi), we can recognize the importance

of tile eciency to make fast tenpai.

• By understanding in what situation to take head start tenpai, we can understand what moves

we need to make to make this possible, which is not the same as maximizing tile acceptance.

(Tile acceptance that leads to a tenpai which it’s better not to take can’t be considered tile

acceptance.)

• By understanding what tenpai is strong (cheap good shape versus expensive bad shape etc.),

we can understand how to balance speed and value in tile eciency before tenpai.

Head start riichi judgment

Head start closed tenpai is usually instant riichi. Please pin this down rst. Then, in what

kind of cases should we go dama (or break tenpai)?

When to go dama

Going riichi or dama will have no eect on nal placement For example in the following

kind of cases:

• In all last, in the lead with a yaku-having tenpai (with yakuless tenpai, riichi is the correct

decision unless falling under one of the cases discussed later).

• In all last, and even with riichi (tsumo or direct hit and ura 1), the nal placement would

not change (for example a pinfu dora 1 tenpai when behind a non-dealer by over 16000, but

being able to secure 2nd place with dama)

• In the round before all last (South 3 in a hanchan), and winning with dama would give us

enough points that no one can overtake us in all last even with haneman tsumo.

• Another player is close to busting out, and tsumo or direct hit on that player would let us

nish 1st.

1.2. HEAD START TENPAI JUDGMENT 45

Situations where the point standing inuences the riichi decision are mostly conned to South 3

onwards and an opponent being close to busting. (Although there are exceptions, the factors below

become more important.)

Dama is expensive enough, so that the point increase from riichi is inecient

• Conrmed dama baiman and up

• Conrmed dama haneman, except with a good wait in the early game

• 3 han 60+ fu hands (example: wertlllzzz777, riichi tsumo and dama

ron are both mangan)

Dama mangan

[long discussion of a simulation and Totsugeki Tōhoku’s book, with the main points being]

• Good shape riichi wins about 50% of the time. Dama only increases this by about 5%, so

riichi is good.

• With bad shape, go dama from the 12th turn onward.

• However, if there is a possible improvement to the bad shape, going dama early is good.

• With bad shape 5200 dama, riichi if there is no possible improvement outside of the late game.

• Riichi good shape has a higher win rate than dama bad shape.

Additionally

• Ryanmen dama mangan

– Dama from the 13th turn (if about 1.5 players are attacking, riichi except just before a

draw)

• Ryanmen dama 6400

– Dama from the 10th turn

• Bad shape dama mangan

– If about 1.5 players are attacking, dama from the 13th turn

– If 1 player, mostly dama

• Bad shape dama 5200

– If about 1.5 players are attacking, dama from the 14th turn

– If 1 player, dama from the 8~9th turn

• Bad shape dama 6400

– If about 1.5 players are attacking, dama from the 11th turn

– If 1 player, always dama (However, bad shape 6400 is often chiitoitsu dora dora or a

san’ankou with improvement to suuankou so dama is already quite probable.)

46 CHAPTER 1. TILE EFFICIENCY

Win rate maximizing damaten

There are some situations where it’s good to dama even medium-scoring hands (good shape 4 han

30 fu and up, bad shape 3 han 40 fu and up) outside of the late game.

Situations where winning is much more important than the added points from riichi

from a point standing perspective

Essentially, this refers to having a decent lead in South. (In East, you usually still want to push

for more points even when in the lead).

Example: South 1 or 2, 1st place and ahead by 12000 points, tenpai for dama mangan

with good shape Getting a dama ron here will create a 20000 point lead, which is safe even

against baiman tsumo or haneman tsumo when dealer. Direct hit on 2nd place creates an even

bigger lead.

Example: South round, 4th place, 1st place has an overwhelming lead, and dama

mangan ron would get us 2nd place

Situations where the win rate dierence between riichi and dama is extremely large

Example: Chiitoitsu dora tanki on a terminal or honor A dicult to use dora will ow

out easily. On the other hand, a middle tile dora will not come out easily, so with a middle dora

tanki riichi is relatively eective. Chiitoi dora dora with a non-dora honor wait is similarly also a

good situation to riichi.

When it’s clear that our target tile is unneeded by at least two other players Concretely,

players that have already cut it, are going for a single suit hand in a dierent color, are going for

kokushi or chanta-type when the target is a middle tile etc.

When our own discard pile is especially easy to read, and it’s clear that our target tile

is unneeded by at least one other player

Defensive damaten

This is the case where we don’t riichi to avoid losing points.

When we’re rst to tenpai, cases like this are quite rare, but they do exist. We will examine a

few situations where we might be prone to hesitate.

All last, 1st place, yakuless tenpai

Unless there are plenty of opportunities to create yaku (including calls that lower value), riichi.

Exception From the mid game on, when we only need to avoid the dealer and have enough of a

lead (at least 4100) or in the endgame when a draw seems likely.

1.2. HEAD START TENPAI JUDGMENT 47

All last, when riichi would lower our rank

Here too, if we’re rst to tenpai, the general rule is head start riichi. However, if any of the other

two opponents are already tenpai, it’s better to dama.

Furiten

For win rate, furiten ryanmen riichi ≈ bad shape riichi < furiten sanmenchan riichi < ryanmen

riichi. (Furiten ryanmen has a slightly lower win rate than bad shape.) However, furiten ryanmen

riichi is always menzen tsumo, so it’s worth more points. Therefore, we should riichi with furiten

ryanmen about as often as with a bad shape. On the other hand, furiten bad shape is extremely

bad.

With an open hand, furiten ryanmen < bad shape. This is because it’s harder to pressure oppo-

nents into folding. If everyone goes betaori, then the disadvantage of furiten essentially disappears.

Therefore, with an open hand, we should take a bad shape that avoids furiten.

Extremely bad wait (less than 2 tiles and not an honor wait)

The so-called ultra bad shape. When calling riichi on the 10th turn, this shape has a 14% win rate

and a 19% deal-in rate, so we should dama everything other than yakuless dora 3. However, on the

6th turn, deal-in rate is high, but win rate is around 30%, so riichi even with 5200 is good.

Example tuaassddvbnm.. dora 4

If y has already been cut twice, it’s not only ri but also etuo. that improve the

wait. When the original hand is bad, the quality and quantity of possible improvements increases.

Accordingly, with an extremely bad shape tenpai from the mid game on, stay dama and try to

improve the shape and riichi, or aim for formal tenpai if the shape stays bad.

Bad shape non-dealer riichi only, in the mid-late game

If we’re dealer and can’t improve the wait easily, we want to riichi and attempt to renchan (a

renchan is worth about 650 points on average).

In the mid-late game, riichi only as non-dealer is sketchy. The expected value is close to zero,

so the riichi decision will depend on the situation. We should especially avoid riichi when leading

in South, when someone has made a threatening call and appears close to tenpai, or when there

are less than 3 draws left.

Wait improvement judgment

Trying to improve the wait (“hand change”) when already in tenpai means we have to delay riichi,

and is basically a loss. It’s better to aim for improvement before tenpai, because there are usually

more ways to improve. Waiting for improvement while in tenpai is only worth it when there are

many tiles that will improve the hand. The following is based partially on simulations by Totsugeki

Tōhoku, which however do not account for the fact that having the initiative over opponents is an

advantage, so the limits have been made stricter.

48 CHAPTER 1. TILE EFFICIENCY

Requirements for hand change

• Early game (4th turn): 6 types to double value, 9 types to improve the shape. (assuming 4

tiles per type, so 2 types of 2 tiles each count as 1 type)

• Mid game (7th turn): Respectively 7 types and 11 types.

• Mid game (10th turn): Respectively 9 types and 11 types.

• Late game (13th turn): Don’t bother. At this stage, it is often better to dama or fold when

the hand is very bad.

These counts include the winning tile.

However, it’s often easier to hand change if we break

tenpai.

There are often situations where a tile will multiply value by 3 or 4. However, with hand change

quantity > quality.

So let’s look at some examples where we would still want to wait for a hand change.

Hand change patterns

A Double serial shape Each serial shape (errt or erty) creates 4 possibilities for

good shape, so 8 in total.

Example rtyuosdfgxxnm,dora 4 → cut o

The point increase is very large, so we hand change even in the mid game. If the nm, was

m,., we only gain pinfu instead of pinfu tanyao, so we riichi in the mid game. If the nm,

was ..., we riichi even in the early game.

B Serial shape + oating tile which is dora or gives yaku

Example qwedghjkzz/m,. dora 4→ cut d

Example qweasdgghjkcc3dora 4→ cut 3and dama to make

sanshoku with c or ittsuu.

C Tanki shape Complex waits and tanki waits are the easiest to hand change. If we have a good

tanki, we can riichi.

Example With chiitoi, hand change to an honor, walled o tile, dora, dicult to use tile etc.

Example wrfffhjkzxcvbndora 4 → cut w

TN: Dama tsumo into furiten riichi is sometimes playable, for example with doraless qwertys

dxcvnnfcutting qfor mangan or haneman. At least that’s what I think he’s referring to. See zeRo’s

book for a discussion of this play (“absolute pinfu damaten”).

1.2. HEAD START TENPAI JUDGMENT 49

D Callable shape Being able to call eectively increases the amount of available hand changes.

Especially common is changing a cheap call into a single suit hand.

Example tyuiooasq

55dora 4→ drop as

E Few types change, but the point dierence is striking These don’t t into the framework

described above, but are still notable.

Example qqqwwwtuiooohj→ drop hj

Example qweaasdjklcb.. → cut b

Example qqqssdffllccc3→ cut 3 and dama.

With sddff or sdfff, riichi.

Exceptional hand change waits

Usually, we want to either riichi immediately, or break tenpai to maximize the probability of a hand

change as a general rule. Outside of late game hands with bad shape and many changes (tanki

etc.) where a draw is likely, or expensive hands (over 5200) with a yaku that have a hand change

to better shape, there are some situations that don’t fall under this.

Cheap hands with yaku, going dama versus breaking tenpai

Example East 1, East seat, 7th turn ertyifgghcvbngdora i

→ cut cand dama

Cutting i and breaking tenpai gives the most hand changes, but because i is dora, dama

tenpai with c gives change on yio (for sanshoku) and a win on u, which is better than the

hand change on zxcvbm on cutting i. The advantage of being able to win on u for

3900 or 2000 all is bigger than hand change in souzu. If i wasn’t dora, cutting it for hand change

would be good.

Example West seat, 5th turn qweerpyggcvbnndora e

→ cut yand dama

Cut n gives hand change on qwrtuxvbm, cut y on r (with tanyao) and

dfghxvbnm. We can’t ron, but the mangan tsumo on gn shifts our preference

to the latter.

When the hand changes in iishanten and tenpai are comparable and the loss from not being

able to win is big, it’s often better to take dama tenpai.

50 CHAPTER 1. TILE EFFICIENCY

When to break tenpai even though dama has many changes

Example qweerttyudfcvbdora b → cut e

The provisional tanki shape (cutting d or f) has the most changes, but cutting e gives

ryanmen tenpai on qwrtuisg and can upgrade to tanyao and/or sanshoku. Cutting

q conrms tanyao, but loses the opportunity of drawing another q for takame sanshoku.

When two triple waits (here qru and wti) remain and there is an upgrade in score

(tanyao, sanshoku), breaking the tenpai excels over provisional tenpai.

Example erttyuuioasdm,→ cut m or ,

Cutting u only gives pinfu, so it’s not worth the loss in speed.

Example werrtydffgghvb→ cut v or b

Taking the tenpai is good, but if we really need mangan, we can cut r.

Wait choice

Which wait to take when in tenpai.

Of course with two equally wide waits, we take the more expensive one, and with two equally

expensive waits, the wider one. With two equivalent waits, we take the one that’s easier to win.

From erpty, we take the wtwait over the eywait as it’s closer to the outside. Between

kanchan and shanpon, we take shanpon. (If the shanpon has one tile less but includes a terminal or

honor, it’s still preferable. If it’s two tiles less, take kanchan unless the shanpon is on a live honor.

However, the read of the table is arguably more important.)

So we’ll consider the situations where one wait is expensive and narrow and the other is cheap

and wide.

With riichi

Point values are for ron with no ippatsu or ura.

• Good shape 1300 > bad shape 2600. This is often with a ety where e is dora or a

ddfggh shape. The tile next to the dora and the iipeikou shape tend to be harder

to win on.

• Good shape 2000 < bad shape 5200. But from qwertyuo cut o for the triple

wait.

• Good shape 2600 > bad shape 5200. Similar to the rst one, but between ryanmen riichi dora

and shanpon riichi dora 2 or 3, choose the latter.

• Good shape 2600 < bad shape 8000.

1.2. HEAD START TENPAI JUDGMENT 51

• Good shape 3900 > bad shape 8000. The gain from ippatsu and ura is maximal at this score.

But from qwerptyuo, cut t for a suji trap.

With calls

• Good shape 1000 > bad shape 2000. Neither has a big enough inuence on placement (it’s

an “interrupting hand”). This is not the same as pinfu only riichi (also 1000 versus 2000),

since pinfu riichi with ippatsu, ura and/or tsumo is worth about 3500 on average, while the

win rate reduction is less than half.

• Good shape 2000 > bad shape 3900.

• Good shape 2000 < bad shape 5200.

• Good shape 2600 < bad shape 5200.

• Good shape 3900 > bad shape 5200.

• Good shape 3900 < bad shape 8000. This gets close in the mid-late game. Also, between

ryanmen yakuhai aka dora and yakuhai aka dora tanki, the former is better because dora is

dicult to ron.

• Good shape 5200 > bad shape 8000.

• Good shape 8000 > bad shape 12000.

With all of the above, if the bad shape is a wait on the dora, the good shape is better.

This can get even more complicated when taking into account takame-yasume situations and

the possibility of going dama.

Takame-yasume

The basic rule is: without any dora, take the takame-yasume. As the number of dora increases, the

preference goes to win rate and conrmed yaku.

Example East round, East seat: weeasdfghcvb11

→ With 0 to 1 dora, shanpon riichi. With 2 dora and up, ryanmen riichi. (If eis dora, shanpon

riichi.)

Example werasdffghjkmm

→ With 0 dora, cut f and riichi. With 2 dora, cut a and riichi. With 1 dora, it’s close,

depending on how likely l appears to come.

Example weeaaadfgxxxmm

→ With 0 dora, it’s sketchy. If the shanpon would be on an honor or we really needed points,

shanpon riichi. Otherwise ryanmen riichi. With 1 dora and up, always ryanmen riichi.

52 CHAPTER 1. TILE EFFICIENCY

Same number of tiles, score versus win rate

Example werfghnnm,,.55, with 5 cut once

→ Take the shanpon riichi, even though it’s slightly cheaper.

Example wweeaaffhzzbb3dora h, 3 is a guest wind

→ Dora tanki riichi

→ With b/, 3 tanki riichi

→ With 3 dora, dora tanki riichi and dora tanki dama are both good

→ With b/ and 3 dora, dora tanki dama is quite good

Example wweeaaff;zzbb3

→ g tanki riichi

→ With any other dora, 3 tanki riichi. (However if the wait is cut twice, riichi on the other

one. Don’t hell wait.)

Case where the wait with less tiles left has a higher win rate

Example qqqrtyudfgjkl3

→ In the very early game (around the 3rd turn) when opponents are not yet done discarding

lone honors, 3 tanki riichi. Otherwise nobetan riichi.

Yasume-takame good shape versus expensive bad shape

Outside of all last where we need a big hand for placement, riichi with the good shape. Good

shape riichi > bad shape dama.

Some exceptions:

Example weasdjklzzzxc

→ When q is dora, riichi makes it very hard to come out.

Example qweadfjklzzzxcdora a

→ Cut f and dama (with a not dora, f riichi is good).

Example weeaaasss..555dora 5

→ Cut w and dama (without dora, riichi).

TN: Heresy. Believe in the ow!

1.3. CALLING EFFICIENCY 53

1.3 Calling eciency

Calling tile logic

So far we’ve discussed closed hands that aim for riichi. We’ll now take a look at hands where we

have called or have the option to call.

Going for closed tenpai every time is a simple kind of mahjong, but it won’t bring home the

bacon. Even with a perfect iishanten like fgxxnnm, it’ll take between 6 and 7 rounds on

average to reach tenpai. Getting a closed tenpai means we got lucky.

That’s why calling eectively is extremely important. Since we’ll need at least 1 han from yaku

to be able to win, we’ll discuss tile eciency that takes yaku into account.

Each call we make can help the shanten advance by 1. Since advancing the shanten is harder

when closer to tenpai, it will often be most eective to call when close to tenpai.

That’s why, with an early game hand that’s far from tenpai, we will more often try to plant

the seeds of a yaku instead of focusing only on tile acceptance. Yaku building is not just

a skill to create big scoring hands. Rather, it’s also eective to bypass the 1 han limit and

increase the win rate.

Now, what are the concrete ways in which tile eciency changes when considering calling?

The value of isolated yakuhai increases

Usually better than a terminal that can’t connect to yaku or dora, but when winning closed looks

tough and we have several single yakuhai, there are cases to prefer them over bad shape joints or

oating middle tiles.

Example qwyysfllccbm56 → cut q, going for yakuhai, open

tanyao or rarely chiitoi.

To the extent that we want to keep these single yakuhai and stack them, it’s standard to

open the hand for yakuhai pon. Conrming a yaku is just as good as advancing a weak

spot. From an eciency standpoint, it only makes sense to ignore a possible yakuhai pon when a

fast and expensive closed hand is in sight.

The preference for paired joints increases

Since we can call both pon and chii.

Example wwwyuidfkkccbm→ cut m

Also, in callable hands, a bad shape joint is preferable to a oating tile since it can be chii’d.

In callable hands, we’ll often want to x a bad shape complex joint that gives yaku and reinforce

elsewhere. In a closed hand, xing a bad shape joint is rare.

Example rrtaasfghklm,,→ cut a

With too much joints, dropping the weakest one and reinforcing the remaining ones is common.

However, there are also cases where we can keep a tile that creates yaku. When going for a single

suit hand, we’ll often drop a joint (rarely, an entire group) of a dierent suit and hold on to several

oating tiles.

54 CHAPTER 1. TILE EFFICIENCY

There are many people who hesitate to call when far from tenpai, but it’s exactly because

the hand is far from tenpai that we must try to call. Cases where we’ll call and then decide

to fold will not be uncommon. If we get into a situation where we can’t win no matter what, it’s not

too late to defend, but there are many point losses that can be avoided with a quick win. Attack

is the best defense.

Calling judgment

With calling, as with any skill in life (?), there are three main skills to know:

1. When to call

2. How to make easy to call shapes

3. How to make yaku for calling to work

When to call

Much can be said about deciding whether to call or not, but the fundamentals are speed and

value. We always have to think of the dierence between the shape after we would call or not

call. We can’t compare a 1000 point bad shape open hand to a riichi pinfu tanyao we couldn’t

possibly have gotten with the hand. If the closed hand would be fast enough and expensive, we can

ignore the calling opportunity, but if it would be slow and not that much more expensive, we’ll call

aggressively.

Example eriisdfhjjnm,dora 3→ chii or pon into tenpai from

the 8th turn on. Yep, even with this delicious mentanpin dora 0. If an opponent looks like he’s

already tenpai or we’re the dealer, it’s good to call even earlier.

With a 1- or 2-shanten hand with a bad shape and a conrmed yaku, we’ll want to call even

more. Yakuhai are usually already 1 call down, so same story.

As a general rule, once we’ve called, it’s no problem to advance the hand with more

calls. It makes the hand faster while not making it cheaper.

Example dfghjxv..4

55→ chii sgk into tenpai

However, calling decreases the number of available tiles, and when we don’t have sucient shape,

it weakens the rest of the hand, making it harder to reach tenpai and easier to end up with a bad

shape. It also becomes harder to integrate yaku and dora. Thanks to these factors, there will be

some exceptional situations where we’ll refuse to call even with an open hand. To say it in other

words, calling further is no problem if we can already concretely envision the 4 groups and the pair.

What about calls where the yaku isn’t conrmed, or leaves the possibility of kata-agari (when

only part of the wait has yaku, a.k.a. atozuke)?

Indeed, just like it’s better to complete bad shapes rst and end up with ryanmen, it’s better

to conrm the yaku rst. But here also, we must compare the speed and value dierence between

calling and not calling. Thoughtlessly hating kata-agari is nonsense. With a hand that would

be too slow to nish closed, call from anywhere; with a dicult to nish hand call only

the weak spots, conrming yaku or eliminating a bad shape.

Example opening hand, qydgkzxx.5677dora t→ insta-call 7

1.3. CALLING EFFICIENCY 55

This is a pretty bad haipai, but if the 7 comes we’ll pon it and cut q. In the distant future,

once our isolated tiles form joints, we’ll be able to call them. If a f comes out after, we’ll ignore

it as the shape has no completed groups yet.

Example 4th turn, wuoghhzz/n577 dora q→ pon 7cut w

We have enough group candidates and pairs, but no group yet. Consequently, the loss in

eciency from calling shrinks. Furthermore, if we don’t call, the hand will likely become too slow.

7 is obvious, but we’d call anything else cutting w, not caring about the obvious atozuke.

Kuikae and kuinobashi

These plays

are commonly hated, but getting extra value or improving to a good shape in 1 turn

is great value. We should proactively use these techniques when calling.

Example Kuikae nashi rules, West seat, dora x

xccvbnm,3344

→ pon 3 and cut c

→ chii . and cut c

→ chii x and cut ,

→ chii / with cv and cut ,

→ chii n and cut c

Planning yaku for calls

Which is the best yaku for calling?

Getting to tenpai closed is possible at most once in every four hands. With most hands, closed is

too slow.

In practice, about half of all wins are open.

Since calling requires a yaku, we must try to somehow create one when a closed win seems

dicult to pull o. Now, what yaku should we prefer for calling with an average hand?

Diagram of open yaku preference

based on ease of making, score and ease of using dora

Kuikae refers to calling chii on ryanmen, then discarding a tile that would also have completed it, for example

cvcut x. It’s illegal in many common rulesets. Kuinobashi refers to any call that doesn’t advance shanten

but still increases score or improves the wait, especially a call while already in tenpai.

As the level of play increases, so does the average tenpai speed. In Houou South, the median speed for declaring

riichi is already on the 8th turn, and open hands are usually even faster! ざわざわ。。。(Source: https://blog.kobalab.

net/entry/20180118/1516202840)

56 CHAPTER 1. TILE EFFICIENCY

Rank Yaku

A+ open tanyao, yakuhai

A honitsu (chinitsu)

B+ toitoi

B sanshoku doujun, ittsuu

B– chanta (junchan)

It’s okay to never win the other open yaku (doukou, honroutou, shousangen, etc.) in an entire

lifetime :^)

If a rare hand where they’re possible appears, we can consider going for them.

Even a low-ranked yaku is a priceless treasure if we can’t see a higher one. Although we

tend to keep chanta and friends at a respectful distance because they’re dicult to make and cheap

when open, we should go for them if we can’t see another yaku. To repeat: we can’t compare them

to a mentanpin that would be impossible anyway.

Prefer conrmed (easy to conrm) yaku

If it’s easy to conrm, it’s easy to make, simple as that. If we can avoid kata-agari, we should go

for it, unless there’s a dierence in score.

Example rpyyii;jv/mm77→ cut 7

Example qqwryfhzxvnm,. → cut h

Cutting w gives the two alternatives of sanshoku and ittsuu, but ittsuu is one tile closer and so

we should cut h according to the “mentor theory”.

Furthermore, if the pair of q was a middle

pair, we should go for the superior pairing of sanshoku and open tanyao and cut z.

Dora

Dora has the best cost-performance of any yaku, but doesn’t count for the 1 han limit. In other

words, once a yaku has been conrmed, there is nothing more powerful than dora. 1 han call

+ dora 3 for mangan is the win with the best cost-performance. With mangan in sight,

we should prioritize speed, but to create mangan and above, it’s no problem to make even quite

painful calls.

Example tyzxv/.. m,

55 dora . → cut x ignoring

honitsu

Example qwerruo

cv5

5dora u → pon rand dora tanki

How to make open yaku

We’ll explain the tricks to making each of them one by one.

maicry.png

It said メンター理論, I have no idea what this means but it sounds cool.

1.3. CALLING EFFICIENCY 57

Tanyao

The indispensable linchpin of calling. An ace that’s fast and can use aka dora easily. Always be

conscious of it.

Example yioofhhjccvbbm→ cut o

When there’s no special dierence in tile acceptance, aim for tanyao right away. It ts closely

with the principle of tile eciency that closer to the inside is better.

Because kuitan is an extremely convenient yaku (especially with many dora), making calls that

don’t advance the shanten but conrm tanyao are seen frequently. Let’s memorize the following

kind of play:

Example weerriihkbm,. dora h→ chii j or n, cut .

Yakuhai

1 han in just 3 tiles, an important yaku together with tanyao.

Keeping single yakuhai in a faraway closed hand is more eective when we have several. The

probability of stack one pair with 567 is bigger than completing a ryanmen (9 tiles versus

8). We can also combine them with other yakuhai, toitoi, honitsu, chanta etc. to increase both

value (especially from 3 to 4 han) and speed. Keeping these single yakuhai over bad shape joints

or oating middle tiles is done to increase our own chances of winning and our score to

the utmost, and not for choking opponents. There are many players who keep these while

narrowing their hand “for some reason”.

Single suit

An excellent yaku that’s comparatively easy to make and can count on a high score. If we have

gathered enough group candidates for the 4 groups, it’s ne to aim for it.

In a closed hand that looks slow and cheap and doesn’t have enough group candidates, we can

sometimes try to force a single suit. In this case, even the humble single guest wind can become

a group candidate. On the other hand, when a hand like riichi + α or yakuhai dora dora is in

sight, we do not approve of discarding tiles in the other suit left and right. Though this can be said

about any yaku, not making an obvious open hand because one is being choked out of useful tiles

is nonsense. No matter how much we get choked, open is still faster than closed. Rather,

we should think of it as slowing down the opponent’s hand. To insist: we can’t compare the open

honitsu to a closed chinitsu that only exists in our dreams.

Toitoi

We often have to choose whether to go for toitoi or chiitoi. In what kind of cases is each of them

good?

When to go toitoi

• We have a closed set

With 4 pairs and 1 set, calling pon and going for toitoi is usually easier to win. Rather than

58 CHAPTER 1. TILE EFFICIENCY

chiitoi 1-shanten for 9 tiles, a callable 1-shanten for 6 tiles is often faster, even when the tiles

are hard to call.

• We can add yakuhai or honitsu for more points

According to the mangan theory, we should go for toitoi even with no set. Toitoi also often

combines with yakuman :^)

• Many easy to call pairs (honors, terminals)

• Needing to win a hand in all last

Calls are great here :^)

• We can go for yakuhai aka or tanyao aka as mainline and make toitoi if we get

lucky

When to go chiitoi

• We don’t have a set

• We have no other yaku Because we can increase the value of chiitoi with riichi.

• We have an isolated dora

• The pairs are hard to call

No problem with chiitoi :^)

• We can also see a group hand

Sometimes happens when we have many iipeikou-type shapes.

• We have a big lead

For defense.

Sanshoku, ittsuu and chanta

We can get these with luck, but we won’t specically aim for these closed unless we really need an

expensive hand. Their cost-performance is not that good, and chanta also prohibits aka dora.

However, in a hand that doesn’t have any other yaku and is full of bad shapes, these three

can be great for getting that 1 extra han. Strong players are uent in taking these hands to the

win. They don’t advance shapes like qwryio closed unreasonably. No way that completes

closed.

We’ll sometimes see hands where several of these yaku are possible at the same time (“balanced

scales”).

Example wasdfjkklzc67

From this bountiful hand we can see pinfu, yakuhai, sanshoku, single suit, ittsuu and chanta.

We can imagine many dierent alternatives as the hand advances.

Even without a conrmed yaku, if the hand is too slow closed, we should see the possibilities of

dierent yaku and try to win using calls liberally.

1.3. CALLING EFFICIENCY 59

Principles of calling

The majority of yaku should be called “whenever possible”

First of all, hands that are at least 5200 with calls, we usually call. With hands that score 3900

but are wide, it’s usually still good to call from anywhere unless it’s the early game and riichi is

realistic. Then what about cheap hands that have the potential to be fairly expensive closed?

Example weryuufgkkcvb dora 4 → call anything from the mid

game on

Even though this perfect iishanten for 6 kinds, 20 tiles can be mangan closed and becomes 1000

open, it’s good to call from the 8th turn on. If an opponent looks close to tenpai, we can call

even earlier. (However, if it’s important to make a big hand to secure placement, there can be

rounds where we’ll ignore calling even late.) With a hand like this that we want to nish closed

if possible, the conclusion is nevertheless to call for tenpai from the mid game on. With hands

that are harder to complete closed (bad shape 1-shanten or ≥2-shanten), it’s usually best to call if

possible. However, that doesn’t mean we can just call anything. If we can’t see the possibility of a

yaku, there’s no choice but to advance closed.

Yaku aren’t just important for score, but also for speed

Since yaku enable us to call, they make the hand as a whole able to accept more tiles. With pon,

we can accept 4× as much tiles, and with chii, 2× as much. (In reality this will be less, because

opponents will often need the same useful tiles and will sometimes even choke us.) Since putting

priority on tile acceptance close to tenpai is the principle of tile eciency, we should try to create

opportunities for yaku in the early game, since the gain in eciency if they pay o is enormous,

especially in a hand that would otherwise be slow.

When deciding what yaku to go for and when to call, decide based on the current

hand, turn and point situation

Of course, winning with an expensive hand is ideal, and ittsuu and chanta are harder to make than

tanyao and yakuhai (and often cheaper when playing with aka). But in mahjong, we can’t always

go for the ideal hand, since 3 times out of 4 an opponent will have a better hand and we’ll have to

cope with a worse one. While chasing the weaker yaku for no reason is to be avoided, if we can’t

see another path to victory, we have to go for them. If the hand is winnable, we should win

it.

Example qeyisfjlzcm.33

With a hand like this, not going for chanta and thinking pinfu is easier to make is a mistake.

Rather, this hand has sucient joints and is hard to win outside of chanta (+ sanshoku if we get

lucky). Furthermore, not wanting to win a 1000 with 2 calls because we’re behind by some points

and persisting for riichi is not good. The probability of reaching closed tenpai fast enough with a

hand like this is extremely low. We should call any of wkx, with this hand immediately and

only keep it closed when behind in all last, or when we have a big lead and don’t care about getting

tsumo’d.

60 CHAPTER 1. TILE EFFICIENCY

There’s not really a need to think about our hand begin easy to read for opponents

Even with a hand like the above example, there’s not really a need to refrain from calling to not let

opponents know what yaku we’re aiming for. Once we conrm a yaku, it’s not really a problem to

get choked since we can call from anywhere. It only makes sense to refrain from calling when the

hand is close to riichi, and then being readable is a minor concern at best. Furthermore, calling

with a faraway hand can put opponents on guard, causing them to weaken their hands (blung).

It might sometimes make sense to not call to not expose an obvious yakuhai atozuke, since those

are very easy to choke, especially refusing a ryanmen chii, since those are the easiest to complete

closed. However, this case will often be connected to the hand being either very slow (don’t want

to call to weaken the shape too much) or very fast (want to riichi instead). It’s ne not to think

about opponents reading our calls.

Example qeuuosfg...77dora f → chii won any turn

Avoiding kata-agari

Obviously, we should avoid it if comes at no cost in speed.

Example With open tanyao, cut w from wet.

Example rtyxcccvb77chii

hk→ cut b

Though we can switch to tanyao if we cut x and later draw n, we want the least amount

of yakuless waits possible.

However, just like with our hand being easy to read, it’s not necessary to consider this an

important factor when deciding to call or not. When we want to not call to avoid kata-agari, it’s

usually because we have insucient group candidates or because we don’t want to call from the

strongest joint rst.

It doesn’t matter that our calls give some opponents more draws

However, it also gives ourselves less draws. If we have the opportunity to call, but the tile is

marginal, and there are many other tiles we can use instead, it’s often better to decline and hope

to tsumo it.

Calling can only complete groups

Now, is it good to call anything with a slow hand that has a guaranteed yaku? It’s not necessarily

always the case. A call might advance the shanten without actually making the hand faster to

complete. We have discussed the 3 ways of advancing shanten earlier:

1. Making a joint from an isolated tile

2. Making a group from a joint

3. Making the head from an isolated tile

With a call, we can only do 2, but by narrowing the hand, we make 1 and 3 more dicult. The

principle of tile eciency is that tile acceptance close to tenpai is most important. With a hand

1.3. CALLING EFFICIENCY 61

that has diculty with 1 and 3, it becomes hard to get to tenpai early and we’ll often end up with

a bad shape. Furthermore, calling makes it harder to integrate extra yaku and dora.

However, in hands where the necessity of performing 1 and 3 is low (in a hand where tanki is

ne, or we can make a nobetan shape like qwer, putting 3 o is ne), in other words in hands

that have sucient group candidates, if we have conrmed yaku, it’s no problem to make one call

after the other. To put it in another way, hands with sucient group candidates can arrive at

tenpai through calls only, without the need to tsumo. Not calling with a hand like that is limited to

hands that are suciently fast and can make riichi (or semi-fast hands where we’ll ignore ryanmen

chii but call the bad shapes), or from the mid game on, hands that are too slow and would have a

low chance of reaching tenpai and calling would make it dicult to defend.

Even with insucient group candidates, it’s ne to call important joints that we would really

like to complete (like yakuhai pairs) and postpone 1 and 3 for later. (We call these kind of joints

weak spots.) Especially calls that conrm a yaku and allow us to call in the distant future are in

agreement with the principle of tile eciency. Even painful calls are ne if they allow us to conrm

a yaku and go for an uncertain but expensive slow hand, especially if the hand would be painful to

nish closed and we can still defend with an open hand.

Even with a riichiable hand, it’s often better to call to improve the shape

Since it’s good to call into tenpai from good shape iishanten with a conrmed yaku from the mid

game on, we’ll often play with an eye to keep the option to call. With a live pair of yakuhai, we’ll

try to play to leave a good shape once we call it, and going for a single suit hand, we’ll drop joints

of the other suits and keep single yakuhai.

Example uiosfhzzxcvb55→ cut s

Seeing pon tenpai, ; or j tsumo. If 55 was aa, we’d cut z. With no ; left, cutting

h for a suji trap is good, preferring the fast tenpai over the upgrade with j.

Example ttuuiicvbnm,66→ cut i

With y, becomes a ryanmen. If 66 was aa, cut a.

There are not many players who excel at calling and the construction of callable hands. (I don’t

consider myself one of them either.)

There are many dierent misguided ways of thinking about calling (however, there’s something

to each “misguided” way of thinking). There’s the yaku supremacists who only make yaku to create

big scoring hands and the menzen supremacists who detest ugly (yakuhai atozuke) calls. There’s

the defensive supremacists who are afraid of narrowing their hand too much with calling. There’s

the riichi supremacists who disdain yaku that are dicult to go for even in situations where going

closed is even more dicult. And there’s the nyaggers who’ll call anything... Indeed, learning the

skill of calling is dicult. But conversely, it’s also a realm where one can show one’s skill dierence

from other players.

62 CHAPTER 1. TILE EFFICIENCY

Dierences from closed tile logic

When calling (or considering it), the value of some joints and isolated tiles changes compared to

closed hands.

The value of pairs rises

Compared to joints that don’t contain pairs, because we can pon twice as much as we can chii.

Even in a 3-pair shape, we’ll often prefer keeping a pair. While compared to a bad shape joint, the

tile acceptance is balanced (2×4 versus 4×2), once we pon, the 2-pair shape that’s left is superior.

We also insure ourselves against shimocha or toimen discarding our bad shape wait. Of course, we

should still prefer a ryanmen over a pair, cutting u from weruuiaddcc

77.

Bad shape joints become better compared to oating middle tiles

Through the inuence of chii, the loss of 8 instead of 4 tiles of acceptance becomes bigger. Even in

a joint overrun where we’d normally prefer a oating tile, we should keep the extra joint.

The value of terminal and honor pairs increases

Because they’re easy to call. In a closed hand, we prefer middle pairs to create ryanmen, but in an

open hand, honors are the best. They’re also easy to fold with if necessary. But when comparing

a 2 to a 7, the ease of calling is similar, but the number of upgrades is double, so it’s common to

prefer the upgrades. Similarly, the easiest to call joints become preferable, but deciding which in

fact is is highly dependent on the discards (especially kamicha’s) so we should pay close attention

to them.

Joints that have a combo become even more valuable

In a closed hand, combos play a role, but usually don’t invert the basic ranking of joints and pairs.

But in an open hand, bad shape joints with multiple combos into ryanmen like sfhjk or

wrtyu become very strong. We should obviously call these to improve them into ryanmen.

This kind of call is called kuinobashi and is easy to overlook, so we want to pay attention to it.

Serial shapes become even more valuable

Chiiing any of etyi into rtyu or any of fgjk into ghhj (still, we’ll only

rarely call h with this shape) creates a ryanmen instantly and is pretty damn strong. While bad

shape joints also become stronger, this kind of shape is so good we should prefer it. Calling from

an insucient group candidate shape is often not very fast, but there’s a big dierence if shapes

like these exist.

1.4. YAKU COMPOSITION TECHNIQUES 63

Aryanmen and oating tiles seperated by 1 from a sequence (“ikken hiraki”) become

more valuable

But not by as much as the above. From eert, we can chii r for ryanmen, from dghj

we can chii f. We can in theory also call any of wety or gk, but the remaining joint

would be bad, and it’s often better to ignore this not very eective call as we can draw a better

tile, but if the turn is late and we want to rush tenpai, we can call. Because it has wider tile

acceptance and it’s easy to make pairs, the aryanmen is a bit better than the ikken hiraki. Their

overall precedence over other isolated tiles is not much higher than in closed hands since only the

chii into ryanmen is actually good.

In the next chapter, we’ll look into general techniques to make specic yaku and use dora,

centering on open hands but also considering closed hands for convenience.

1.4 Yaku composition techniques

Principles of yaku and dora

The most important things to remember about yaku and dora. In Japanese mahjong with riichi

and its copious amount of dora, making an expensive hand without any yaku is easy. Accordingly,

techniques pertaining to yaku are useful less frequently and have less of an impact on play than

those pertaining to basic hand eciency and dora usage. Therefore, when looking at the hand,

instead of rst thinking about what yaku to make, we should rst think about where to make

groups and whether going for riichi is realistic. (In Chinese mahjong, where riichi doesn’t exist but

a yaku is still mandatory to win, it makes sense to think about yaku rst.) That’s why in this

strategy guide, we rst discussed tile eciency focused on “where to make groups”.

If creating yaku or using dora is the most ecient move, we should play it

This really goes without saying. When comparing equally ecient (fast) alternatives, we should

always prefer the any that gives dora or yaku. The problem is when yaku and dora are not optimally

ecient, or when we have to compare them to each other. This chapter will center on such cases.

The value of dora and yaku changes depending on the existence of other dora and

yaku in the hand, and is therefore relative

In mahjong, each han up to mangan doubles the value of the hand, but any han above that are

inecient. Accordingly, when we have no other dora or yaku, we’ll consider them important, but

when the hand is already a conrmed mangan, the added value of extra dora or yaku is small.

(There will of course be cases where the point situation will make going for haneman and above

important, but in the majority of situations, going for the fastest possible mangan is good.)

Aim for easy to conrm yaku

If both are equally easy, aim for the most expensive one (not just han value, but stackability).

This the principle of comparing yaku. Unless the score gap is big, easy to conrm yaku are

preferable. This can also be said about comparing dora to yaku or yaku to speed. Chasing too

hard to get yaku is rare, and so is breaking down yaku that have already been created. With

64 CHAPTER 1. TILE EFFICIENCY

“conrmed”, for global yaku (tanyao, honitsu, toitoi) we mean that all the group candidates have

been gathered. For partial yaku (yakuhai, iipeikou, sanshoku), we usually won’t break them up if

the groups have completed (except in hands like qwasdghzxcmmm, calling riichi

when we draw q or w). Furthermore, hands where multiple mutually exclusive yaku are both

reachable (“balanced scales”, ryōtenbin) are often preferable, unless a wider or more expensive

shape exists.

Play to make dora and yaku easy to conrm

For example, when there are multiple ways to call a tile, we should expose the group that conrms

dora or yaku. From errty where e is dora, we should chii for

er to leave rty

which can slide to accept another dora. Same idea without dora ebut when chasing 345 sanshoku,

because there is a small chance we might draw y and prefer to cut e for defense. By exposing

the group which gives yaku, we make sure it won’t get destroyed.

Often, dora is more important than yaku

Because it’s easier to use. When deciding what yaku to aim for, the ease of using dora will often

be a deciding factor. But when we have no other yaku, yaku is more important than dora

because it allows us to call. In slow hands that are expensive even without using a particular

dora, it’s also better to go for a yaku that enables us to win.

When deciding whether to chase yaku, it’s not just important how close it is, but

how close another way of winning is

For example, since sanshoku requires 3 groups, in a hand with 2 groups already complete, we’ll

almost never go for it unless desperate for points. Conversely, when the hand is far from tenpai, it

makes sense to consider even far away yaku. Keeping isolated tiles that create value, and dropping

other joints when a tile sticks to them, all the time without decreasing shanten, is called watari-

uchi. In mahjong, since tenpai speed is the most important of all, the skill to chase yaku without

straining (slowing down) the hand is important.

On the other hand, in a situation with insucient group candidates, forcing a certain yaku by

dropping a joint (kime-uchi) is usually not done, even if the rest of the hand is strong. However, we’ll

see some exceptions to this rule when going for tanyao and honitsu, two yaku that are exceptionally

strong and allow us to call freely.

It’s most important to be good at making common yaku

This chapter will therefore focus heavily on the most common ones. For rare yaku, we can deduce

the kind of logical moves by analogy with more common yaku or by applying general rules.

Example In a far away hand, keep a in qqzza for sanshoku.

Example Pinfu iipeikou riichi > bad shape ryanpeikou

Example Ryanmen honitsu yakuhai 2 > tanki honitsu shousangen. Without honitsu,

we usually take shousangen.

1.4. YAKU COMPOSITION TECHNIQUES 65

Example With 2 kans down, we should make a 3rd one, but “going for” sankantsu is impossible.

Having discussed generally when to go for what kind of yaku, we’ll give a brief sketch of when

to go for each one in particular and the special moves for chasing them.

Tanyao

Extremely easy to make and callable without kuisagari, this yaku forms the foundation of mahjong

together with riichi.

Isolated tanyao tiles become more valuable

In a hand where tanyao is achievable, oating 2/8 >> single yakuhai. Even if tanyao isn’t

conrmed, if we have no other yaku, playing towards the center is standard.

23 and 78 ryanmen become much weaker (obvious)

If we can conrm tanyao, we usually should

Example weryudfgvvnn.. → cut .

The exception is hands like weryudfgcbnn.. where to conrm tanyao we

have to sacrice a ryanmen riichi, similarly cutting t from wet even if the rest of the hand is

all tanyao.

Even if it lowers shanten, we can drop a penchan and keep a oating middle if it

conrms tanyao

Since we can call, it doesn’t just increase the score, but also makes the hand faster.

Example qwtuusssfhkvvb→ cut q

We can call without advancing shanten to conrm tanyao

Example weerr;jklccn,→ chii any of hm

Example qwertysfkkcn,→ chii any of ruand cut q

But with qwertyfgkknm4, we’ll prefer riichi pinfu, unless leading in all last

where score doesn’t matter.

Similarly, when score isn’t an issue, the following kind of big brain call is easy to overlook:

With the rest of the hand being zxcvgjj, cutting z after the chii:

errtty→ chii any of ru → left with errt

rrttyu→ chii any of ey → left with rtyu

wrrtty→ kanchan chii e → left with rtty

66 CHAPTER 1. TILE EFFICIENCY

Pinfu

This is also the foundation of mahjong, but rarely something to aim for, more often just a bonus

for playing eciently.

Because pinfu means we have minimal fu, it’s not as good for scoring as the other 1 han yaku.

It’s good to not worry about losing pinfu with an otherwise good play. It’s common to prefer

another benet over pinfu.

Example qertyfgghjvv44→ cut q

Temporarily sacrices pinfu, but the upgrades into good shape are more plentiful (honor tile

shanpon is also strong). If we draw the w, we cut 4 aiming for pinfu tanyao despite furiten.

However, with ≥5 han, fu no longer matters. It’s helpful to remember that pinfu tsumo + 4

han is one of the main lines to haneman.

Since a yakuhai pair invalidates pinfu, in hands where we have enough sequences and ryanmen

if we don’t have a head, guest winds become slightly better than yakuhai.

When pinfu is realistic, we’ll sometimes x a ryanmen pair into a ryanmen more often than

eciency alone would suggest. We’ll also prefer ryankan over penchan/kanchan pair.

Example wrtyyuasdllnnm→ cut n

Identical acceptance to cutting w, but the ryanmen kanchan (remember?) guarantees pinfu.

Example wryyisdfghzzm,→ cut y

Cutting w or i when the next group completes.

Example xccvbnmm,,ghll→ cut ,

This creates a double ryanmen kanchan waiting on any of zvn. (remember?)

Iipeikou

Like pinfu, iipeikou is another yaku that is rarely specically aimed for but comes naturally as a

reward for ecient play. The shapes that give rise to iipeikou like aryanmen and nakabukure are

naturally easy to extend and should be preferred over regular oating tiles.

Since iipeikou is a 1 han yaku only, when all other joints are ryanmen, we’ll drop a iipeikou-

giving bad shape joint rst. (As discussed earlier, good shape n han > bad shape n + 1 han.) Since

we already hold one tile of the wait, if another one gets discarded we end up with a painful ultra

bad shape. In this case, it’s often good to prefer a bad shape joint with 4 tiles left. (Again, a 2×

dierence in speed > 2× dierence in value.)

We can often aim for iipeikou with a “ying pairs” shape, creating two groups from it. This is

easy to overlook, so we should pay attention to it.

Example wertyuffggkkbm→ cut m

The tile acceptance is 2 tiles lower, but we cut m anyway (not b as drawing v is big here).

Making it easier to end up with ryanmen is an important factor too (pinfu, easier to win).

Example wertyussffhhxx→ cut x

1.4. YAKU COMPOSITION TECHNIQUES 67

Example qqeyyiidgxvm,. → cut x

This is slightly inecient (3 pairs is bad) but leaves the tiles for iipeikou. If xv was cb

(inner kanchan), cutting y would be good (not i as we want to draw t for ryanmen).

wweeerrr weeerrrt

The two complex waits above also give iipeikou on w or t. This can be very easy to overlook.

Yakuhai pairs

Giving 1 han (or 2 han with double yakuhai, renfonpai) with just 3 tiles, yakuhai have an excellent

cost-performance value. This makes single yakuhai comparable to isolated number tiles, even though

they’re vastly harder to make a group with. There will even be hands where they are preferable to

joints. For this reason, the use of single yakuhai is extremely dicult and forms a major obstacle

to any systematization of tile eciency. We’d like to acknowledge that this section will therefore be

longer than those on other yaku and more theoretical. We’ll start with the more straightforward

discussion of paired yakuhai.

When to drop a yakuhai pair

Since we can call pon, a 2 pair shape of which one pair is yakuhai has an eective tile acceptance of

10 tiles (4×2 + 2). Though this will be slightly lower in practice as we’ll not always be able to call,

it’s still comparable to ryanmen. The honor shanpon when in tenpai also has a win rate rivaling

ryanmen. Therefore, it only makes sense to drop a yakuhai pair in the following kind of cases.

Conrmed pinfu iishanten But even here, if the advantage of keeping yakuhai is large, we

should keep it.

Example errtp;hjklzz77dora k → cut z

A mangan even when called, and the shape after the call is very wide and guarantees a good

wait. When aiming for haneman and up, cut 7.

Example qqwweefgkkvb66dora 4 → cut f

Preserves the option of going for chiitoi. The fg has a very slight overlap with kk, and

we can draw ; for chiitoi.

We can make tanyao without increasing shanten

We can cross into a superior yaku From honitsu into chinitsu etc.

Example xccvbnnm,,,.66dora . → cut c

But with a hand like this which is already expensive enough, and the dierence is speed is large,

it’s better to take the straightforward win.

Conversely, there are also situations where dropping a ryanmen to keep a yakuhai pair is good.

68 CHAPTER 1. TILE EFFICIENCY

Example weyuifghxc5566dora 4→ drop one of the ryanmen

Example qqerrtyb/jkl77dora q → cut r

Taking tenpai even with w.

Should we 1-call yakuhai pon?

(By 1-call, we mean to open the hand and make the rst call of the hand.) Yes.

The exception is the following kind of cases:

• We can expect a fast closed tenpai, for example an early game iishanten with 2×ryanmen

and a yakuhai pair, and calling would be cheap. We should especially not call if the closed

tenpai would be expensive.

• Chiitoi iishanten (but if we can combine with another yaku for expensive toitoi, we can

call)

• The hand is so fucked it’s better to go for kokushi musou

• Not fucked enough to go kokushi, but the hand has no head, group or dora and

it’s already mid game, and we’re in the lead.

If the hand is very far away from tenpai, we should still 1-call in the early game. In the

future, once we tsumo some joints, the advantage of being able to call will be very big and have a

large impact on win rate. It’s not very important to think about defense in the early game. If the

tsumos are bad, we can start emphasizing defense from the mid game on.

Should we call for yakuhai atozuke?

When we have gathered enough group candidates

Example qeoonm,,77

ghdora 7→ kamicha discarded w

If we chii it, our atozuke’s wait will become extremely obvious. Should we call? Our other

options to make a less obvious to read tenpai are w or 7 tsumo or 7 pon. 2 or 3 tile types is

not nearly enough to wait for a hand change, and we should always chii.

Example wwexxxmm55

dfdora 5→ kamicha discarded q

Totsugeki Tōhoku wrote in his Scientic Mahjong

that we should chii, but should we?

We’ll apply the same theory as in the previous example. We get a less obvious shape with

qwrm5 tsumo, wm5 pon and r chii. (wm tsumo also open up sanankou, but this is

so rare of a benet it doesn’t really matter.) Depending on how we estimate the ease of calling

these, this is about 8 to 10 tile types, so we should chii from the mid game on.

Example oocv.66qq

ldora 6→ kamicha discarded x

The famous and highly technical book which contributed signicantly to the popularization of digital mahjong.

It rst came out in 2004 and is now (2019) in its 22nd printing.

1.4. YAKU COMPOSITION TECHNIQUES 69

Adding to xb6 tsumo and 6 pon for a better wait, we have zm, tsumo for chanta,

cvtsumo for toitoi and the ultra spicy . tsumo for toitoi honroutou with a chance of sanshoku.

We should also ignore this chii in the early game and call from the mid game on, unless we really

want to score big. Furthermore, if o comes out, we should pon to assert dominance and cut v.

Because even in extreme cases like these, calling is good from the mid game on, it’s no problem

to call into yakuhai atozuke in cases where our yaku isn’t obvious to opponents even from a high

shanten. It’s really necessary to worry about getting choked.

Even when we could get choked, it’s more important to consider the turn and the

quantity and quality of improvements to the hand.

When we don’t have enough group candidates

We should obviously call the yakuhai, but calling the rest of the hand won’t always make it

faster. It’s good to only call weak spots that would otherwise be especially hard to complete, like

joints around the dora or with an overlap. Our judgment will often depend on the quality of the

entire hand and the turn.

Example 1st turn, qyasslcb,5677 dora t→ only call 7

Example 5th turn, wrybmsfhjkl77dora 4→ only call 7or n

From the mid game on, call the rest too. (But g makes it harder to complete the head, so it’s

sketchy.)

It will also often depend on the discards and the opponents’ speed.

Another thing to pay attention to when calling for yakuhai atozuke is that the yakuhai can’t

become the head.

Example rrtvvbsdf77

n,→ cut t or b

Preserving the 2 head candidate shape. Pay attention to not cut r or v in this kind of

shape.

Single yakuhai

The value of single yakuhai

Usually, we think that 1 < single yakuhai < 2 < single renfonpai, but the inuence of the rest

of the hand is big. The value of single yakuhai increases when the hand is dicult to win (bigger

inuence on win rate) and when they lead to a strong hand (honitsu, toitoi). They also have the

advantage of often being safe when outpaced by opponents. Even with 1 tile discarded, we want to

keep them if a closed win looks dicult.

Example 1st turn, qwyiiddgjlxv56 dora t→ cut q

Even if we draw e quickly, the hand is still too slow and cheap, so we’ll try to go for yakuhai

or open tanyao.

70 CHAPTER 1. TILE EFFICIENCY

Example 1st turn, erassg;klvbn57 dora 4→ cut 5 or 7

Even though the penchan shape is bad, this hand is already 2-shanten.

Example 1st turn, ertiiaskkl2257 dora 2→ dicult

Already 2-shanten, but with many bad shapes and expensive even open. There’s also a possibility

of going manzu honitsu. Whether to drop as, cut an honor or cut kas a compromise is dicult.

Single yakuhai discard order

Because it’s good to deny an opponent the renfonpai (especially double East is scary), cut round

wind → dragons → seat wind. Though this is highly debatable, from guest winds it’s good to

cut shimocha → toimen → kamicha. We want to prevent shimocha calling chii on our discards,

while being able to chii from kamicha’s fast hand is an advantage. There’s also the advantage or

disadvantage of extra draws. But when there’s specic opponent we want to prevent from a quick

win, it’s best to cut his wind rst.

How to keep single yakuhai

Single yakuhai become better when we have several Because it’s easier to stack at least

one. In a slow hand with 3 dierent ones that has many bad shapes and no other yaku, we should

drop the weakest joint and keep the yakuhai.

Furthermore, when playing like this, oating middle tile > penchan. However, we’ll still

prefer a ryankan or inner kanchan over a single yakuhai, hoping to luck into drawing a sequence

rather than more yakuhai.

It’s often good to keep single yakuhai in faraway hands with too many joints

Single yakuhai become more valuable when we already have a yakuhai pair or set

Because it’s easier to go for honitsu and/or toitoi. We’ll discuss this in the respective section on

those yaku.

Sanshoku

Formerly known as the ower of the yaku, today we more often hear “don’t force sanshoku” and

“speed over sanshoku”. Both old style players who chase for it too much and players who overem-

phasize speed even when they should go for sanshoku abound. (The same can be said for many

other ≥2 han yaku.) We’ll try to explain using ideas from tile eciency when to see sanshoku and

when to disregard it.

If we can see sanshoku without a real loss, we should obviously see it. When comparing oating

tiles and joints of the same type, or when discarding e or y from erty, it’s important to

verify the possibility of sanshoku.

Example wrydfgcvvbn.. → draw h cut d, draw s cut g

Example qwadhjkllzxcvb→ cut v

1.4. YAKU COMPOSITION TECHNIQUES 71

When comparing riichi pinfu and bad shape riichi sanshoku, the latter is preferable, especially

in a hand like the example where sanshoku is conrmed and pinfu isn’t. (We cut v over b to be

able to accept /.) The score is higher, and since we can chii (and we should on any turn unless

we need a big hand, since an iishanten with 2 bad shapes takes an average of 17 turns to tenpai),

the speed is comparable.

If qw were we, sanshoku is not conrmed but pinfu is, so dropping ad is good, going

for sanshoku only when needing a big hand and having no dora.

Example wertyasdfhllvn→ cut n

We can make conrmed 456 sanshoku, but the 3-sided wait and the amount of upgrades to

ryanmen make going for pinfu a bit better.

In general, bad shape joints that conrm a 2 han yaku are preferable to ryanmen, especially

when far from tenpai since they enable calling.

Example wryuiohhklvb,. → cut w

Similarly, isolated tiles that can create sanshoku also become stronger.

Example qwetyuasggzm56→ cut 5 or 6

Since making sanshoku requires 3 groups, when we have 2 strong group candidates outside of

the possible sanshoku, we disregard the possibility of sanshoku from isolated tiles. (The same idea

holds for ittsuu.)

Example qqetttyuhcvbnm→ cut h

The cvand qqettshapes are too strong to justify keeping hfor uncertain sanshoku.

While this example doesn’t amount to it, returning shanten to chase sanshoku is almost never done.

However, with too much joints, the loss from keeping an isolated tile for sanshoku while dropping

the weakest joint is small, even when being able to call.

Example wwrtuuioghbm22→ cut 2

Taking a shape that doesn’t conrm sanshoku but leaves two alternatives is also often good.

Example ertyffglllvb,,→ cut f

This move foresees both 345 and 456, and is preferable to keeping the vertical acceptance for

riichi only. If l were dora, the added score from sanshoku becomes weak and it’s better to cut e

and maximize speed.

Example wertyuughhvvbn→ cut v

While cutting h enables both 456 and 567, the dierence in tile acceptance is very large and

so this move isn’t recommended. Cutting v for a very wide shape and a shot at 456 is better.

When considering WWYD problems about sanshoku, the following criteria will usually lead to

the correct answer:

• Ease of conrming. In general, a sanshoku from an isolated tile is hard to conrm, while

sanshoku from a bad shape joint is easier, but people often seem to prefer the former.

72 CHAPTER 1. TILE EFFICIENCY

• When hard to conrm, go for sanshoku when the loss in tile acceptance is small. When

close to tenpai, this loss becomes relatively bigger and sanshoku more dicult to make, while

far away from tenpai the advantage of making sanshoku is big. Old style yaku supremacists

tend to neglect the former, riichi supremacists the latter.

• In fast hands where the benet of being able to call is small, go for sanshoku when the

marginal value of 2 han is large. With a conrmed mangan, sanshoku usually doesn’t

matter, but with no other yaku or dora, it’s very big. Here too, yaku supremacists tend to

neglect the former and riichi supremacists the latter.

We’ll now consider a few example problems based on these ideas.

Example weertuijklbbm,→ with no dora cut e, else i

Sanshoku is not conrmed (requires double takame), but the dierence in tile acceptance is only

4 tiles.

Example pydffhjzzzxcnm→ cut hj

Sanshoku is not conrmed at all (requires triple takame), so we cut the weakest joint since we

have too much of them.

Example errtyuigghjnmm→ cut n

Requires both of ,k for sanshoku, and the shape on cutting n is much wider.

Example tyyyuid;hjjknm→ cut i

While we can cut d and go for both 567 and 678, losing out on mangan tenpai with f is

painful. Cutting i for the easier to conrm 567 with b is better.

Example wrryuiojzzcbm,→ cut w

At this stage, cutting j which gives 678 or 789 is a no-go. The w move which preserves

2-shanten and a strong 2 pair shape is good.

Example tuiissdfgjccbm→ cut t

This hand is far from tenpai, but since we’ll commonly go for open tanyao making 567 closed

will be rare. Unless we really need an expensive hand, cut t.

Ittsuu

The theory of ittsuu is similar to that of sanshoku. When 2 good group candidates outside of

ittsuu have been gathered, we ignore it, but we prefer bad shape joints that guarantee ittsuu over

ryanmen.

Example qwrtyiollvbbnm→ cut vb

Example qwtyuiollvbbnm→ cut qw

1.4. YAKU COMPOSITION TECHNIQUES 73

Similarly, an isolated tile that can create ittsuu is stronger. The q in qrtyio is

obvious, but even the a in the uncertain afgkl shape is easy to overlook.

Example qwryuiofgcbm,,→ cut q

In this example, ittsuu is conrmed, but cutting qo for (open) tanyao is more eective. Same

as with sanshoku, it’s often better to not chase it too much when tanyao is available.

While sanshoku and ittsuu are mutually exclusive, there will be some shapes where both are

achievable depending on draws.

Example qwetuiodzxcvmm→ cut v

Example ertyiidxcvbnm,→ cut d

The above example has a big dierence in tile acceptance, so it’s better to go for ittsuu or

tanyao.

When we have to pick between sanshoku and ittsuu, we’ll think about the following criteria. In

general, it’s better to pick the more exible sanshoku which can stack with tanyao.

• Ease of conrming

• Strength of acceptance

• Score depending on other yaku or dora

We’ll examine some example problems of this type now.

Example East round, North seat, 7th turn, dora ,

eeuisdghjklm,. → cut gfor sanshoku

This is an iishanten with conrmed pinfu. Sanshoku and ittsuu both give the same score, so we

only need to nd out which one’s easier to make.

To make ittsuu, we need to drop uiand draw both a and fin either order. While drawing

f rst gives a 3-sided wait, we still only get ittsuu on a.

On the other hand, dropping gh for sanshoku only requires drawing o. If we draw it rst,

we can enter tenpai with conrmed sanshoku. Since sanshoku is easier to conrm, we should go for

it over ittsuu.

When going for sanshoku, we could also drop sdinstead of gh. However, the af wait is

usually easier to win. Also, cutting gand drawing n later allows us to shift over to 678 sanshoku,

which can stack with tanyao.

Example East round, West seat, 8th turn, dora o

weasdfgklzxcmm→ cut kl for sanshoku

Here, making sanshoku and ittsuu is about equally dicult, but going for sanshoku guarantees

pinfu.

74 CHAPTER 1. TILE EFFICIENCY

Example East round, West seat, 8th turn, dora d

weasdfghjlxcmm→ cut l for sanshoku

At rst sight, ittsuu is conrmed and sanshoku looks dicult to make. But even here, going for

the more exible sanshoku shape is better.

When going for ittsuu, we need to drop either of the ryanmen. If we draw k rst, we get pinfu

ittsuu, but if we draw the ryanmen rst, we get ittsuu with a bad wait.

On the other hand, when we cut l, no matter which of qrzv we draw, we can make

pinfu takame sanshoku, since we can slide in pinzu to make either 123 or 234. Furthemore, with

both r and v we get tanyao.

Toitoi

We usually start calling for toitoi when we have 5 sets or pairs. (Obviously, with 6 pairs we should

take chiitoi tenpai.) With no other yaku, calling for toitoi from 4 blocks or less is painful, but if we

can combine with another yaku like yakuhai, tanyao or honitsu, this is no longer the case.

Toitoi or chiitoi?

When to go toitoi

• We have a closed set

With 4 pairs and 1 set, calling pon and going for toitoi is usually easier to win. Rather than

chiitoi 1-shanten for 9 tiles, a callable 1-shanten for 6 tiles is often faster, even when the tiles

are hard to call.

• We can add yakuhai or honitsu for more points

Especially when we can still make a decent open hand even without toitoi. We can also hold

one extra tile in a non-toitoi hand and transition to it when we draw a pair.

• Many easy to call pairs (honors, terminals) Even better when we can call a middle tile

to start.

• Needing to win a hand in all last

Calls are great here.

• We can go for yakuhai aka or tanyao aka as mainline and make toitoi if we get

lucky

• Suuankou iishanten In general, we should call into mangan tsumo tenpai immediately from

this shape and only force suuankou if behind by a lot.

In modern meta, it’s common to force toitoi from 4 pairs 0 sets if it’s still the early game, there’s at least one

other yaku, and most of the pairs are easy to call. (look up トイトイダッシュ for examples and more strategy)

1.4. YAKU COMPOSITION TECHNIQUES 75

When to go chiitoi

• We have a pair that has been cut twice and can no longer become a set

In this case, we’ll often break a set and force chiitoi.

• We don’t have a set

• We have no other yaku Because we can increase the value of chiitoi with riichi.

• We have an isolated dora

• The pairs are hard to call

• We can also see a group hand with iipeikou

• We want to play defensively

When in the lead, or there is an especially dangerous tile (daisangen threats etc.)

• 2 sets, 4 pairs Breaking a set creates an extremely wide iishanten that guarantees chiitoi

tenpai on almost any possible draw. But if circumstances for toitoi are good (yakuhai pair,

early turn), we can attempt toitoi.

Example problems

Example yyiiassdzznnm,dora n → cut ,

Whether to cut any of adm, for chiitoi or s for toitoi.

With chiitoi, the hand accepts 9 tiles for 6400 dama. With s, the hand accepts 6 tiles (yiz)

for 2600 riichi and 4 tiles (u) for 5200 riichi.

While cutting s allows an upgrade to ryanmen with any of tbnm, including p/, it’s

still not as good as chiitoi. Also, dropping the dora onto the table constitutes a risk. We should

cut the least likely of adm, to be left in the wall, but in the absence of information, cutting

, allows us to cope with drawing /n.

Example yyuughkkbnnm,,dora , → cut y

This problem isn’t easy. In practice, the two alternatives are y and g.

Cutting y emphasizes win rate. Calling for 3900 is the mainline, but adding aka, sanshoku or

dora pon makes it easy to reach mangan. Going for 567 sanshoku closed is unexpectedly problematic

as it requires letting go of a dora, so we should call any of pk/m, from the early game.

Cutting g preserves chiitoi iishanten with an outlook for a monster hand up to ryanpeikou.

While either is a correct answer, we recommend y. The main advantage of g is preserving

chiitoi, but the wait becomes bad, and calling a lucky dora or aka makes the yyuu shape

heavy. We think a 3900 minimum with good shape is the main idea.

76 CHAPTER 1. TILE EFFICIENCY

Example eeruuiiodf,,77dora 4 → cut df

Cutting , is fastest, aiming for 7 pon into perfect iishanten. This kind of fast move that

ignores chiitoi is good when leading in all last. But it’s very cheap. Cutting 7 is the choice of the

menzen riichi faction, but it’s quite slow and not much more expensive, so it’s a bit lacking.

Dropping df is a move that aims for maximum upgrades into an expensive chiitoi or toitoi

hand. After 7 pon, it’s faster than cutting , and allows for a natural climb into toitoi.

While r preserves the chiitoi iishanten, forcing the df shape is a bit awkward, and breaking

up the er ryanmen makes the iishanten worse, making this the least skillful move.

Especially when playing without aka, ignoring toitoi/chiitoi with yakuhai and no dora makes

the hand very cheap. Dropping a ryanmen to preserve these yaku is an indispensable strategy in

aka nashi rules. In aka ari, toitoi and chiitoi become weaker because they have trouble using aka,

but even there df is the best move.

Example eeeiiaagg,4566dora 4 → cut ,

With 1 set and 4 pairs, whether to keep the set or widen the chiitoi acceptance depends on the

hand. Keep the set when

1. there is a yakuhai pair or tanyao

2. there are no dead pairs

3. the pairs are easy to call (honors and terminals)

and force chiitoi when

1. there’s a oating dora (especially an honor dora)

2. toitoi would be 2600

3. there are many middle pairs

is the general theory.

In this dicult problem, whether to go for chiitoi or toitoi depends on personal preference, but

having polled some strong players, cutting , and calling from anywhere seems to be popular.

While the oating dora is cause for concern, the chiitoi acceptance is only 3 tiles lower, and toitoi

hatsu is often easier to win than chiitoi dora tanki. But since toitoi has trouble defending, if

defensive power is important, it makes sense to cut e. Toitoi is more proper to attack, as is chiitoi

to defend.

Example eeeiiaagg,,,66dora 4

This is either a very wide iishanten for chiitoi, or an iishanten for toitoi up to suuankou.

Here too, toitoi is preferable. While chiitoi gets to tenpai faster, it’s not that easy to draw a

good tile to wait on. It’s better to pon into tenpai quickly.

1.4. YAKU COMPOSITION TECHNIQUES 77

The problem is whether to ignore pon and aim for suuankou. In online mahjong without special

prizes for yakuman, it’s recommended to pon from the 6th turn on. Suuankou is the easiest yakuman

to make, but drawing one of the 6 needed tiles and then tsumoing on of the remaining 4 before

it comes out from an opponent is dicult, even more after skipping a possible pon. Since we can

tsumo sanankou after calling, we think it’s good to call here.

Chiitoitsu

For deciding between toitoi and chiitoit, see the previous subsection. This subsection will focus on

deciding between chiitoi and a sequence hand.

Chiitoi or sequences?

When far from tenpai, not deciding for either and preserving both possibilities is usually good.

However, there are some exceptional cases where committing towards one shape is better. This

is typically in hands where one of the two is dicult to achieve and can be safely ignored without

a major loss, while the committing move signicantly improves the win rate of the more realistic

alternative.

Example qqrdgjjzzcbb,. → cut ,.

This hand is already 2-shanten for chiitoi, while making a sequence hand is very slow. Breaking

up ,. for chiitoi preserves a small chance of making 345 sanshoku.

In a hand like this with at least 4 pairs where the chiitoi shanten is lower than the group shanten,

and no possible draw can create a good group hand, committing to chiitoi is good, holding on to

isolated tiles that look easy to draw or easy to wait on. While committing moves that pointlessly

negate yaku just lower win rate, this kind of committing move improves speed close to tenpai and

is therefore in accordance with the principle of tile eciency. This isn’t just true for chiitoi, but

also for yaku like single suit or chanta, where breaking up a joint to keep a yaku-related oating

tile is often good.

However, if there are sucient joints for a group hand and we have a yaku that enables calling,

it’s often better to break up pairs and go for a normal hand. The loss from missing out on chiitoi

will be small since we’ll want to call in the majority of cases anyway.

When close to tenpai

The basic idea is to commit to groups when the group shanten is lower, make a mixed shape when

the shanten is the same, and preserve chiitoi shanten while leaving the option for a group hand

when the chiitoi shanten is lower.

Example werrtuiiokkxvv→ cut i

Example wwtuuffggxcvvb→ cut t

While a group hand with tanyao is very realistic, breaking up the y to preserve chiitoi is

actually faster. If tuu were yuu, lowering shanten and cutting f to commit to pinfu

tanyao is good.

78 CHAPTER 1. TILE EFFICIENCY

Example eetjkllzxxbbm. → cut z

The group shape is not strong enough to commit to it, so we preserve chiitoi 2-shanten.

Example pyysdffhkxxvbb→ cut y

With tanyao, ending up with chiitoi will be rare, and the kmove that preserves chiitoi weakens

the group shape by a lot. Fixing the aka ryanmen is better.

Example wwerrtffhhjjvb→ cut vb

Preserving chiitoi 1-shanten while envisioning up to ryanpeikou with gk.

Example wwerrtffhhjjv/→ cut h

With the aka in souzu, breaking up the souzu would be painful. We cut hseeing 456 sanshoku.

Example wwwerrffhhjjvb→ cut vb or j

Here dropping vb is also good, but if f or h is dora, aiming for open tanyao with j is

faster, creating a good shot at perfect iishanten with the next advance.

In the three examples above, we shouldn’t cut manzu as this would narrow the tile acceptance

towards a group hand too much.

Example wwrffgccvb1155dora r, East round → cut c

Since we can and should call from anywhere, we can ignore chiitoi and cut c. If 1 were a

guest wind, we’d cut g preserving chiitoi. We’d still call 5, but the other calls would depend on

the circumstances.

Example errdfxxccvbbnn→ cut nearly and efrom the mid

game on

When preserving 1-shanten is important, cut e, but returning to 2-shanten with n creates a

better shape in the long term.

We should return a chiitoi 1-shanten into a group 2-shanten in the following cases:

• The loss from taking a mixed shape would be big.

• We can call from anywhere or we can easily return to group 1-shanten by drawing one of

many ryanmen.

• The turn isn’t too late. From the 10th turn on, it’s good to preserve any kind of shanten,

even if it’s narrow.

Single suit hands (honitsu and chinitsu)

Good cost-performance that can be expensive even open, it’s common to force these when far from

tenpai. That’s why it’s often hard to judge whether to go for single suit or not. In this subsection,

we’ll focus on hands where such a decision is dicult and try to establish some basic rules of thumb.

1.4. YAKU COMPOSITION TECHNIQUES 79

When to commit to single suit

We have at least 10 tiles of the desired suit, including honors that have been cut less

than twice, and at least 4 blocks

We have no groups in another suit Unless a multi-suit hand would be cheap with a bad

shape, and the single suit hand would be expensive and easy to call.

Example werzzxxvvn,... → drop manzu

We have at least two strong blocks in the o suits With one o suit block, a single suit

hand can still be made very naturally preserving shanten along the way, but with 2, going for single

suit is often overdoing it.

Example qweruuioooasxc→ cut u

We’d rather go for junchan sanshoku here, even though we have 10 manzu.

We have another open yaku (especially yakuhai) but less than 3 dora When not going

for single suit gives at least 3900 or mangan, not going for it is common, but we still should if speed

isn’t aected too much.

Example 6th turn, dora g

rtdfgghjkl155

Since we get 3900 without single suit, cutting 1 is good. But when we draw d or h later,

we’d have to cut dora for 2000 tenpai, so at that point we’d revert to haneman iishanten by cutting

from manzu. If the manzu were a bad shape, we’d drop them immediately in the example situation.

If a hand satises all four of these criteria, going for single suit mainline is no problem, and

we should call everything to advance. (Except in hands like qwerttyuui444

which are very wide and would be too cheap open.)

However, these criteria are for committing to single suit. If the hand doesn’t satisfy one of

them, we should still play with an eye for single suit, hoping for good draws. How much to go for

single suit in hands like that is more dicult. Let’s call a hand state that can commit to single suit

A, and a hand state that can safely ignore single suit B.

Consider the relative ease of achieving A and B

When we have many tiles of the single suit and one o suit joint, we get to B when we tsumo

into the osuit joint, and we get to A when we tsumo or call into any other block, which will

also create a callable shape that can advance faster. In this kind of case, A will usually be much

easier to achieve than B, and it will often be good to drop the the o suit joint while holding on

to single suit oating tiles and live honors, calling into a single suit even without sucient blocks.

In a hand where the relative chances are more equal, it’s often good to cut isolated honors without

overcommitting to either A or B.

To quote zeRo さん: if we have a yakuhai pair, we should always take 2 seconds to think about honitsu.

80 CHAPTER 1. TILE EFFICIENCY

While the relative ease of A and B depends a great deal on the shape of the numerous suit, what

many players fail to realize is that it also depends greatly on the shape of the osuit. Especially in

a bad early game hand, it’s quite common to commit to even an uncertain single suit if no other

yaku are in sight.

Consider the relative strength of A and B

Simply put, this refers to their relative score. With 2 yakuhai pairs (or 1 yakuhai pair and 2 or 3

single yakuhai) or when holding a dora or aka in the single suit, it’s easy to make mangan and we

should call even more aggressively. On the other hand, when we already can make mangan even

without single suit, or when we have an osuit dora or aka, it’s best to go for yakuhai + α and

maybe single suit if we get good tsumos. We should especially keep single yakuhai when going for

single suit is realistic.

Example West seat, 2nd turn

yhllcbbn,22445→ cut y or h

Honitsu mainline, with toitoi as a backup plan. It’ll probably end up as honitsu only 2600 but

since it’s dicult to win otherwise we can’t help it.

Example 4th turn

etuaadffgj1337→ drop manzu

If we can make one more block in pinzu or honors, we’re set.

Example 6th turn

yassdfghjjlbn6→ cut 6

We have 10 pinzu, but drawing onto the osuit is realistic and we can make other yaku, so we

don’t commit.

Example 1st turn

qwfhkxcbbm,567→ cut qw

Choosing between yakuhai and penchan. If we can stack yakuhai, we can see mangan, so we

want to keep this option open.

Example 8th turn

wesdhk,,135577→ cut we

We have enough blocks for honitsu mangan, and we can also make toitoi mangan. Because it’s

a late turn, we shouldn’t overcommit to either, but depending on the progress of the opponents we

can settle for 1000 or 2000.

Single suit hands are powerful, but they are also easy to read for opponents. However, it’s best

not to worry about this too much. Making an expensive hand without opponents noticing is rare,

while going for a single suit when the hand is bad can scare opponents and is actually an advantage.

However, when we have completed the blocks for the single suit, it can be good to cut honors and

tiles from the single suit to camouage. Also, when committing to single suit, it can be good to

discard tiles from the second most numerous suit rst to make opponents uncertain about which

suit we’re aiming for, making it slightly easier to call.

Example yxcbm,11133

55→ cut b

1.4. YAKU COMPOSITION TECHNIQUES 81

Since single suit hands often give rise to many-sided waits, it’s good to know these (see subsection

on many-sided waits). http://hinakin.main.jp/mckonweb/index.htm is also a good trainer for

chinitsu (warning: 20 second timer, HARD).

Mahjong veterans’ single suit criteria

Though we’ve discussed some criteria of single suit commitment judgment, there will be many cases

in actual play where deciding will be dicult, and play styles will dier even among strong players.

We have therefore chosen to include the following blog posts (needs an account to view) by チルノ

さん and 氷室さん.

From チルノさん’s “幽雅に咲かせ、無理染めの桜”:

I’m not really a single suit kind of player, but I’ll try to nd out the border of life of single suit,

with normal hands in an equal point situation. In all examples, assume East round, South seat,

2nd turn, all yakuhai live.

2 yakuhai pairs and up (see blog for other hands, I might translate this in the far future)

Example rio

9999999

5577

If the other suits are all bad shape, this is about the border. If the isolated honors are dead, I

have an osuit ryanmen or a dora-related joint etc. I’d ignore single suit. Haipai with 2 yakuhai

pairs usually means going for honitsu or toitoi.

Example wryi

999999

5577

Even with an osuit ryanmen or dora-related bad shape, this is about the border of single suit.

If I have an osuit group or several good shapes, I’d try to balance.

For example:

Example riodgzxx135577dora j → cut z

This could well become toitoi, but start with z.

Example riogjzxx135577dora j → cut 3

Comest mangan is when we draw a second j. The z feels useless but I’ll start with honors.

Example rioassbn135577dora j → 2000 is ne I guess

Example rioasvb1235577dora j → cut as

Maybe won’t honitsu but starting with as has a good feel.

Example riosgjvn135577dora b → normal mode

Example wriosfvb135577dora m → cut sI guess

Example wruissvb135577dora m → cut vb?

Let them bloom nobly, the cherry blossoms of forced honitsu

82 CHAPTER 1. TILE EFFICIENCY

Example wriossfvb35577dora m → perhaps not single suit

Example rtuuiossvb5577dora m → cut s

Might drop s here, but it’s hard...

Example wriossf,.35577dora m → honitsu looks good

Example wriosscbm35577dora m → not in manzu

Maybe in souzu though.

Example wriosfjl135577dora m → look at discards

Probably look at discards and see which honitsu looks easier to make.

Example qtu

9999999

5577

Toitoi or honitsu with a pair from a dierent suit looks better, doesn’t it?

Example rt

99999999

5577

Looks like something else will stick before honitsu.

For example:

Example qtusgcc1345577dora z → cut q

Example qtusgcc1345577dora i → sketchy

Example qtusgcc1345577dora e → cut s

Example wtudhcc1345577dora m → cut w

Example wtudhcc1345577dora i → sketchy

Example wtudhcc1345577dora e → cut h

Example wtuahcc1345577dora a → start with h?

Example rtdhcc13465577dora j → cut d

Example rtfhcc13465577dora m→ sketchy, doesn’t look like

honitsu

Example rtghcc13465577dora y→ doesn’t look like honitsu

Example rtghzc13465577dora y → nope

Example rtfhzc13465577dora y → cutting z feels okay

Example ryu

9999999

5577dora r

With one more dora this gets good enough so don’t want to force too much. Yakuhai 3 dora 1

also exists so single yakuhai become important.

1.4. YAKU COMPOSITION TECHNIQUES 83

Example ryudgjzx135577dora r

This is still zx but

Example ryuggxc1235577dora r

Example ryuggjvn135577dora r

hands like these are 3.

Example wryuggjv,15577dora r

Don’t commit to honitsu but cut ,.

Example ryuiggjv,15577dora r

Same but still ,→v→1.

From 氷室’s blog:

1. When the dora is an honor or in the suit we’re considering Honitsu suddenly becomes

much stronger, and getting 3900–8000 is easy. Being aggressive and making the opponents

wary is not bad. Even with a riichi to the face, holding out with some honor tiles is common.

(By the way, when the pressure is strong like this, opponents won’t riichi shitty hands, so you

should pay attention.)

2. When a closed advance would give a shitty wait for 1300–2600 but honitsu might

give 2000–5200 The so-called forced honitsu from a shit hand. Honitsu stacks easily with

other yaku, and can be freely called so it will even complete once in a while :^) However, with

a bad haipai, 1000–2000 yakuhai isn’t bad at all. There are many players who try to force

honitsu to the bitter end without considering clearly faster opponents or the wider situation.

The same can be said about defense. Being no-ten with a 2000 honitsu with 3 calls down

should be avoided Especially when opponents can obviously see how cheap the hand is.

3. When we have 2 yakuhai (or 1 yakuhai and some live honors) Gotta go for honitsu,

toitoi, dora and so on for at least 3900–8000. Getting 2000 with the rare double yakuhai

starting hand is a tad wasteful.

4. When we need at least 3900 at any cost, and the hand is shit This occasionally

happens in all last, and forcing honitsu there is quite good. (For some reason many people

hesitate to do this.) With a shit hand, getting 3900 closed is hard :^)

And here are some special traits of my own style of going for honitsu:

1. Focus on score over shape This might be unexpected, but I don’t care that much about

how many tiles of the suit I have. Of course, that’s doesn’t mean I don’t think about it at

all... How high can I score with honitsu? How high can I score without honitsu? are more

important questions when I’m deciding.

84 CHAPTER 1. TILE EFFICIENCY

2. Play condently When going honitsu is clearly better, it’s normal to drop osuit tiles and

keep single honors. This isn’t really a special trait... but when spectating I see a lot of people

who undervalue single honors.

3. Floating honors This ties in with the previous point, but when holding several of them,

I often think of what the shape will look like when one of them pairs up. This isn’t really

supported by statistics or anything, but with 3 of them drawing a pair happens pretty often.

4. When the shape or score is meh, don’t overcall and keep safe tiles By the way, this

isn’t just for honitsu. But it’s a pretty dicult skill I guess. You might miss the train, but

you don’t want to get run over by it :^) Especially with forced toitoi it’s even easier to die.

Chanta, junchan, honroutou

Similar to single suit in that it restricts the entire hand, but harder to make and worth less points.

With a yakuhai pair and no dora going for honitsu gives 3900 or 5200, but with chanta it’s only

2000 or 2600, which is a big dierence. There will be many cases where a cheap hand which can be

won comfortably will sacrice speed for honitsu, but rather than going for chanta it’s often better

to go for yakuhai only. The value of chanta only becomes apparent in mangan-class hands like

yakuhai chanta dora 2 or junchan sanshoku dora.

That being said, if winning closed looks unrealistic, and we can go for a reasonable open chanta,

we should. In general, go for chanta when we have at least 4 blocks. We shouldn’t restrict

ourselves to a closed hand, but call from anywhere.

Chanta is easy to combine with sanshoku, and there will also be many cases where we can go

for either chanta or ittsuu. The idea for comparing the two alternatives is similar to comparing

sanshoku and ittsuu (see above).

Example East round, South seat, 8th turn, dora .

weadfhjklm,.44→ cut h

Going for chanta/pinfu good shape over ittsuu bad shape, this is a fairly straightforward prob-

lem.

Honroutou can often be made by upgrading a chanta or toitoi, so it’s not very important to pay

attention to it specically.

Example 1st turn, dora 4, 3 is a guest wind

qeyiadjlxvm.33→ cut y or v

We’ve treated this example before. We cut yor v (to preserve sanshoku) and call everything.

If in the West seat, we cut the same tiles to begin, but don’t force chanta, going for West only and

maybe chanta.

Example 1st turn, dora 4

qeyuasjlcvm.44→ cut y or v

Even with ryanmen we still go for open chanta. But in the West seat, it’s faster to drop as

and go for yakuhai.

1.4. YAKU COMPOSITION TECHNIQUES 85

Example 5th turn, dora s

qqeuoasdxcnm.. → cut n

At this stage, riichi is not dicult and junchan isn’t conrmed, but we can call anything except

v. While the loss on drawing v hurts, on drawing anything else next going for junchan is both

faster and more expensive.

Sanankou

While it can be scored with an open hand, it’s a closed yaku in practice as the 3 sets must be

entirely self-drawn. As it’s dicult to make and undervalued at 2 han, the necessity of being aware

of it is low.

Example wwuiossflllvvn→ cut f with no dora, else v

With a mediocre no yaku no dora hand, we should preserve the weak 3 pair shape if it can make

sanankou. With any yaku or dora, or if vvn were vvb, it’s good to cut v.

Example rtyyyuukkzzzcv→ cut r

While u is widest, preserving iipeikou and sanankou gives a big dierence in value. If we have

two closed sets before tenpai, it’s indeed good to be aware of sanankou.

While comrmed sanankou can sometimes be upgraded to suuankou, if we have no other dora

or yaku, the riichi increase from 3200 to 6400 is very big and the standard move is instant riichi.

With dama 6400 we can wait for an upgrade.

Kokushi musou

With the exception of the coincidental yaku tenhou and chiihou, all the other yakuman except

kokushi musou can be made by upgrading a normal hand (daisangen from shousangen, chuuren

from closed chinitsu, the rest from honitsu or toitoi). Since we can aim for these yakuman by going

for a lower hand and crossing over into them if we get lucky and draw the right tiles, and their

appearance rate is extremely low, we won’t explain any related techniques separately.

However, since kokushi musou requires us to completely give up any option of another hand,

we’ll consider it shortly even though it’s rare.

Should we declare a draw?

Simulation results for kokushi chasing

Starting hand draw rate win rate deal-in rate opponent tsumo rate

13 types, 13 tiles 0% 99.812% 0.057% 0.048%

12 types, 13 tiles 2.834% 66.542% 10.637% 9.859%

12 types, 12 tiles 3.797% 55.718% 14.052% 13.162%

11 types, 12 tiles 11.049% 20.960% 23.977% 22.067%

11 types, 11 tiles 11.074% 19.749% 24.525% 22.381%

10 types, 11 tiles 15.024% 6.464% 27.919% 25.453%

86 CHAPTER 1. TILE EFFICIENCY

From these stats, we see clearly that going for kokushi from 11 is good. From 10, it’s

sketchy. From 9, declaring a draw is usually good. Of course, this depends on the situation,

but even when safe from last place by mangan tsumo, declaring a draw from 9 is usually better.

When the scores are especially close, it can be good to declare a draw from 10, even when trailing.

If a hand with 9 honors and terminals looks reasonably winnable as yakuhai, honitsu, chanta,

honroutou etc. for about mangan, we can continue, taking these yaku as mainline unless we draw

the 10th tile for kokushi very quickly. With multiple yakuhai pairs, it’s dicult to commit to

cutting a middle tile, so it’s better to ignore kokushi and call from anywhere for yakuhai 2 + α.

When going for kokushi mainline, it’s ideal to keep a backup plan with one of the above yaku or

chiitoitsu for when one of the necessary tiles dies, but it’s often not possible to preserve this option.

Unless we can go for another yaku early on, it’s better to go for kokushi.

Example quzzxxc.123456→ cut u

When committing to kokushi, we cut the most dangerous tiles rst, holding on to safe tiles. Since

opponents will be able to notice the kokushi no matter what, we usually don’t worry about disguise,

but when starting with 10 types, 12 tiles, we can cut an unneeded tile early on to camouage.

Nagashi mangan

The opposite of kokushi. When we have many honors and terminals, going for kokushi is usually

better from a score and defense view, and we’ll usually only consider nagashi in the late game when

going for a normal hand but getting all bad draws. When we have enough honors and terminals for

guaranteed nagashi, we should go for the win even if they are dangerous towards an opponent in

tenpai. This is because a guaranteed nagashi mangan has a higher win rate than even a very good

wait. If an opponent is going for nagashi, we shouldn’t forget to call and break it up. Similarly,

when committing to nagashi ourselves, we should cut the easier to call tiles rst.

Dora in closed hands

Dora is amazing for value, and unless we need to emphasize speed in a situation where score doesn’t

matter, or when defense is especially important, we usually only want to cut dora close to tenpai.

Of course, any blocks and oating tiles that can integrate dora become more valuable.

Since the value of dora depends greatly on the rest of the hand (if the hand is already mangan,

the added value of dora is low, if we have conrmed tanyao, a terminal dora becomes fairly useless

etc.), comparing dora to other components is very dicult. It’s easy to make a WWYD problem

by taking the most ecient discard and turning it into the dora. We’ll leave the more dicult

problems (1–2-shanten eciency) for later and focus here on the easier to compare cases.

• Comparing components of the same class

• Assuming at most one other dora and at most pinfu as yaku, so each dora doubles the score

Sliding sequences

With dora w, we should cut y from erty to accept the dora.

1.4. YAKU COMPOSITION TECHNIQUES 87

Ryanmen

We should of course keep dora ryanmen. Since there are 4 omote dora but only 1 aka dora, we

should prefer a dora-accepting ryanmen to an aka-accepting one.

At 1-shanten, we commonly break up the overlap in eryufg, but if t is dora and we

have no other dora or yaku except pinfu, we can sometimes break up fg and aim for pinfu dora

2. The dierence in tile acceptance is only 4 tiles, but the dierence in score is very large. (With

1 han elsewhere, it’s better to go for speed taking riichi, tsumo, ura dora into account.)

Bad shape joints

If the dierence is only 1 han, we prefer a ryanmen. When comparing bad shape joints to each

other, the ones with dora are much more preferable, and the 1 han dierence is far more important

than the subtle variations in shape that we discussed earlier. With an overlap into dora, we should

remove it unless we have no other dora or yaku as above.

Complex joints

We usually prefer to x into the shape with the most dora, even if inecient.

Isolated tiles

Being able to draw a second dora for a dora pair is far stronger than being able to make a ryanmen.

Of course, being able to make a dora ryanmen is also very good. In the general ranking of isolated

tiles based on ease of making ryanmen, which is honors and terminals < 2·8 < middle tile < serial

shape, an isolated dora is one rank higher.

While dora neighbors are also stronger, it’s not enough to rank them up against more ecient

tiles. (Because a ryanmen is better than a bad shape with dora. But when comparing tiles of the

same class, we especially prefer those next to the dora.) Of course, 1 tile away from dora > 2

tiles away from dora.

When holding the dora, the tiles 3 away (dora suji) becomes less valuable. After cutting w

from wp and drawing e, the loss is smaller. It’s not 1 rank worse, but from qts; it’s

about bad enough to cut s. The tile 4 away also becomes slightly weaker, as making a ryankan

with the dora at one side is not as advantageous. We cut a from qta;.

But when we don’t have the dora, the dora suji becomes more valuable. With dora r, we

should cut q from qa. If we cut q, we might draw w or e next, cut them, and draw dora

later. It’s a very subtle dierence, but worth paying attention to since it’s easy to discard the dora

suji thinking it will become dangerous in the future.

Dora in open hands

If we have already conrmed an open yaku which can’t use the dora, we should of course get rid of

it rst, since it will be dangerous in the late game. But if we can draw in the vicinity of the dora to

88 CHAPTER 1. TILE EFFICIENCY

create a good closed hand or a dierent yaku, we should hold on to it by cutting a less useful tile

rst. In open hands, dora is even more important than in open hands for the following reasons:

• Open hands have more control over completing groups than closed hands, so it’s

easier to use dora.

• Closed hands can use riichi, menzen tsumo, ippatsu, ura dora to score high with-

out really paying close attention to it, while open hands can’t use these yaku,

making dora relatively more valuable.

Yakuhai 2 becomes 8000 with honitsu or dora 2, quadrupling value. On the other hand, pinfu

riichi has the same face value of 2000, but with tsumo, ippatsu and ura actually averages around

3500. Adding sanshoku or dora 2 to it only increases average value to 9000, which is only by a

factor of 2.5, and adding value is also more dicult than with an open hand. The dierence is big.

There are many players who don’t realize this and play with a xed image of closed = big win,

open = cheap and quick. There are many hands where this will be true, but from the viewpoint

of expected value, it’s often important to emphasize dora and yaku with an open hand.

Please remember this.

An isolated dora becomes especially good when we have a yaku and enough blocks, allowing

us to stack dora or aim for dora tanki tenpai. This is extremely good for both speed and score.

(Especially when we have exactly 2 han elsewhere.)

Example weruajjkv/466dora a

With a hand like this, it’s good to call from anywhere, aiming for a mangan tenpai like the

below. While this hand has trouble defending, it’s not that bad to push against an opponent’s

riichi, so emphasizing speed is no problem.

Example wera

v/66

jdora a

Example psddhkxxv,,.4dora t

With a hand like this we should call anything for open tanyao.

Example p

xx,,

hkdora t

This is indeed a bit overkill, but making a sequence in manzu and drawing another aka is another

way to mangan.

Atoatozuke

With conrmed yaku and enough blocks, we should usually call anything unless we’re condent in

a better draw (depending on the turn and the speed of the opponents). With no conrmed yaku,

it’s still often good to conrm a group and try to create yaku later on (atozuke). We’ve talked

about these, but there’s a surprising number of cases where calling is good.

For example, if we have only a single yakuhai, we can still consider calling to complete a group

and try to draw a yakuhai pair later. Since this kind of calling is for an even less conrmed yaku

than atozuke, we call it atoatozuke.

Of course, atoatozuke is not a good idea as often as atozuke. We can say it’s eective in the

following kind of cases:

1.4. YAKU COMPOSITION TECHNIQUES 89

• The hand is slow and a closed tenpai would be very dicult. (With a reasonable riichi

hand, calling like this would only make the hand harder to win.)

• The completion degree of yaku is low, but we can see multiple possible yaku to oset

this.

• Close 2nd in all last, a hand with dora 2 etc. where winning is guaranteed to be high

scoring or wining is especially important.

For possible yaku, we need to see at least 3 or 4. We also want to avoid moves or calls that

make some of these yaku become mutually exclusive (especially with open tanyao). Also, since the

actual chance of winning is low, we should remember to consider defense from the mid game on.

Concretely, we mean the following kind of yaku:

• Yakuhai (1 yaku for each single yakuhai)

• Tanyao

• Single suit (with 3 blocks)

• Toitoi (with 3 pairs)

• Chanta (with 3 blocks)

• Sanshoku (needs 2 more tiles to conrm)

• Ittsuu (ditto)

All of these examples are starting hands.

Example qwryiagllvn57→ chii t, cut a

Ittsuu, 456 sanshoku, yakuhai.

Example epooc/agkl567→ chii rv, cut a

345 sanshoku, tanyao, yakuhai.

Example tiaadgjkm,..5→ chii l., cut t

Junchan, 789 sanshoku, ittsuu, yakuhai.

Example (2 is a guest wind)

qqwyooxn.2256→ pon qo2, cut n

Honitsu mainline and toitoi, chanta, honroutou, yakuhai atoatozuke.

Formal tenpai

Since no-ten payments exist, it’s sometimes good to call even with no yaku (formal tenpai, keiten).

If opponents aren’t explicitly attacking, call from a good shape 1-shanten with about 16

tiles left and from a bad 1-shanten with about 20 tiles left is a good criterion. When tenpai

is especially important, like when trailing badly in the dealer seat with a bad hand, it’s good to

call into keiten even earlier, but when no-ten payments wouldn’t change placement but winning

90 CHAPTER 1. TILE EFFICIENCY

would, it’s good to try and win to the bitter end. Situations also exist where going for keiten with

an upgrade to yaku is possible, and we should also call a bit earlier then.

In keiten, we can’t win except for haitei, but we can still deal in, so we want the least amount

of draws possible. Just before a draw, we should call if it allows us to avoid drawing a tile while

keeping tile and dealing a safe tile.

In the next section, we’ll approach tile eciency from yet another perspective, centering on

complicated hands, especially those in 1-shanten.

1.5 Iishanten eciency

The importance of being rst to tenpai

The diculty of comparing two alternatives of diering natures

When deciding between a ryanmen and a kanchan, it’s always better to choose the ryanmen unless

there’s something special going on, so that there’s nothing to hesitate about. The reason is that

when comparing things of the same nature (joints), we can decide based on tile acceptance alone.

Conversely, we only waver when comparing two alternatives which dier by their very nature.

In mahjong, there are two great choices between things of a dierent nature. The rst is

between value (yaku and dora) and speed (tile acceptance, wait quality). The second is between

hand components of dierent types (isolated tiles versus extra tiles in complex joints etc.). Though

we have discussed both of these comparisons to some degree, when these factors enter the equation

in a more complicated way, things can get very dicult. It’s especially troublesome to choose the

correct discard when there are many eective tiles.

Luckily, we can often deal with hands that are far from tenpai by applying the rules that

have been briey discussed until now about comparing isolated tiles and joints depending on their

surroundings and the rest of the hand. The need for complicated comparisons is restricted to 1–2-

shanten hands. As we will see, there is a limited amount of possible iishanten types that will make

it easier to systematize this kind of decisions. (That being said, 2-shanten will be the stage where

we’ll most often have problems deciding.)

The advantage of head start tenpai

While we’ve already pointed out that emphasizing tile acceptance close to tenpai is a fundamental

rule, this is because being in tenpai (especially as the rst player) is a big advantage. Formerly,

strategies that put the most emphasis on tile acceptance when in tenpai, that is wait quality, were

the mainstream. The reason this strategy has changed is the following.

Until a few years ago,

getting into tenpai with a good shape (ryanmen or better) was considered

especially important. This tendency can even be felt in Totsugeki Tōhoku’s older posts. Since the

step from tenpai to winning on average takes the most time of all, it was thought that aiming for a

good shape tenpai even at the expense of tenpai speed increases win rate. This is also the reason

why bad shape riichi was often unfairly undervalued.

This is a simplied presentation of a very dicult subject. Entire books could be written about formal tenpai

(look up 形テンの極意 by ASAPIN).

In general, things of the same nature have the same qualities but in diering degrees insofar as they are similar,

making it easy to choose that alternative which exhibits their common desirable qualities to a greater degree.

roughly 2000–2005

1.5. IISHANTEN EFFICIENCY 91

However, because a hand with yaku can win from anywhere (a hand with riichi can be hard

to ron, but by making opponents fold or roll, it increases the time available to win relative to

opponents), the #1 longest step is from closed iishanten to closed tenpai, because we can

only use the tiles we draw. (The actual win rate of bad shape riichi was also shown to be not that

bad by statistical analysis.)

Accordingly, the strategy of maximizing tile acceptance at iishanten and instantly call-

ing riichi even with a bad wait has recently become mainstream.

Of course, there are hands where it’s better to prefer making a good end shape in iishanten.

Together with the trade-o between speed and value, hands like these will form the subject of this

section.

Example qweydffglllb44

This hand is widest on cutting f (3 tiles more compared to y and b), but in practice this

is never played, preferring the higher chance of a good shape and the score from possible iipeikou.

Classication of iishanten

These are the basic types of iishanten. Since they dier by their inherent nature and have dierent

advantages and disadvantages, it’s often dicult to decide which of them to go for when several

are available. For this reason, it’s important to be aware of each of these patterns to be able to

recognize them quickly.

(1) 2 groups, 2 joints, 1 head: extra tile shape

Example qwesdjkcvb..4

Characteristics: since the extra tile doesn’t directly contribute to tile acceptance, the total tile

acceptance is often low. However, the extra tile can be used as a safe tile, or to upgrade into a joint

that makes for a more valuable hand.

(2) 2 groups, 1 complex joint, 1 simple joint, 1 head: no extra tile shape

Example qwessdjkcvb..

This shape consisting of a ryanmen and a ryanmen pair is called perfect iishanten.

Example qwedddfjkcvbn

This shape with a many-sided wait is also considered to fall under (2).

Characteristics: often wider than (1). Only hands like the second example can end up with a

many-sided wait that isn’t sanmenchan.

(3) 3 groups, 2 joints (or 1 joint and 2 isolated tiles), no head: headless shape

Example qweasdjkcvbm,

For a gentle introduction to the subject, see https://youtu.be/mKEOEWEc5JE (EN subs available)

92 CHAPTER 1. TILE EFFICIENCY

This shape has 2 joints, but the weaker shape with j3 instead of jk also belongs in this

class.

Characteristics: since we can get into tenpai when one of the joints turns into a group or into

the head, it’s wider than (2) if we have 2 joints. However, if we don’t have a serial shape or closed

set, we get a tanki bad shape if we draw a group.

Example qweasdjk333m,

This shape is called the anko headless. When we complete a ryanmen, we can downgrade the

closed set into the head and end up with the other ryanmen as the end shape. Please remember

this very strong shape.

(4) 3 groups, 0 joints, 1 head, 2 isolated tiles: sticky shape

Example qweasdjcvbm..

Characteristics: since turning an isolated tile into a joint is much easier than turning a joint

into a group, this shape is the widest of all. However, it can be easy to get a bad wait, unless we

have a serial shape.

(5) 5 pairs: chiitoitsu iishanten

(6) Combination of (5) and (1) or (2)

(7) Kokushi musou iishanten

For these, see the subsections on chiitoitsu

and kokushi musou.

Quantitative comparison of classes

Table of tile acceptances for common iishanten shapes

acceptance into good end shape/total acceptance

iishanten class 2× bad bad + good 2× good good + ultra good

extra tile 0/8 4/12 16/16 19/19

perfect 0/12 4/16 or 8/16 20/20 23/23

headless 0/20 6/24 12/28 26/37

anko headless 0/20 10/24 28/28 37/37

honor terminal 2 or 8 middle 3445 3456

sticky 0/4 0/12 4/16 8/20 14/17 14/29

Includes half the 0/2 acceptance for drawing a set from the pair, so just take the sum of the two

isolated tiles, for example qwertyuusm accepts 4/16 + 8/20 = 12/36 tiles.

The material on chiitoi in this book is quite thin. I’ll try to remedy this once I’m done.

This table wasn’t present in the original, I hope this will convince people more clearly that wider shapes are in

fact strong and by how much

1.5. IISHANTEN EFFICIENCY 93

Extra tile shape

In a shape with 2 groups, 1 head and 3 joints, we have to decide which joint to break. As discussed

before, we should use the criteria for comparing joints. If there is no dierence in tile acceptance

or value, we should drop the joint whose backres have an overlap.

Example qwetysshjxcvbnm→ cut y

Drawing r upgrades the hand to tanyao.

However, when comparing simple joints in 1-shanten, there’s one thing to keep in mind. Al-

though more central joints have more upgrades, more outside joints are easier to win with, so we

need to balance the relative frequency of cases where we draw an upgrade and those where we

complete a group. When comparing an outer and inner kanchan, the win rate disparity is not that

big, so we should prefer the inner kanchan’s double upgrade. But when comparing a pair of 2s

and a pair of honors, the latter is signicantly easier to win with, so we’ll usually prefer it. The

opponents’ discards also play a large role. When not in 1-shanten, drawing an upgrade will happen

a lot, so we should generally prefer upgrades.

When deciding which extra tile to keep from a situation other than the 6-block one above, we

should keep an extra tile that is most valuable in some way.

What should be kept in mind is that at 1-shanten, an upgrade that overlaps with another wait

can’t be considered an upgrade and we can only perform it by refusing tenpai, so we should ignore

it unless declining tenpai is for some reason very advantageous.

Example yuiooadcbm,→ drop cb

While the cb theoretically has more upgrades than ad, drawing n already puts the hand

in tenpai. Accordingly, choosing the easier to win s wait is better.

What to keep as the extra tile

There are several dierent ends towards which the extra tile can be used:

• A tile which is useless but serves to confuse opponents (rarely, if no better option is available)

• Safe tile

• Floating tile to make a ryanmen

• Floating tile next to the head (can advance into a wider headless or sticky shape if the sequence

completes)

• Tile to create value from yaku or dora

In the early and mid game in a normal situation, the options are ranked from worst to best.

Dealing in with the extra tile doesn’t happen that often so we can emphasize our own hand, and

upgrading to double value is often preferable to upgrading the end shape. When both joints are

already good, keeping a safe tile is better than a tile which creates ryanmen or a tile which can

advance the hand to a wider iishanten class (unless this would be very advantageous somehow). If

tiles that create value appear dangerous from the mid-late game on, we should also get rid of them

94 CHAPTER 1. TILE EFFICIENCY

rst and keep a safe tile. In a shape like wep where the dora can be immediately used to enter

tenpai with a good draw, we usually don’t get rid of it.

With a bad shape, we don’t keep a safe tile. Since the dierence in win rate with an eective

extra tile is signicant, increasing the deal-in rate a little still has better expected value. Since

getting outpaced with a hand like that will usually result in folding completely anyway, dealing in

with the extra tile will happen rarely.

The comparison of extra tiles can be done using the material discussed above. (When keeping

a safe tile, keep the safest one, or one that’s safe against a player we especially don’t want to deal

into from a points standing perspective.) It’s easy to disregard this kind of comparison as opposed

to comparison of joints, so we should pay attention to this.

Example yuoosdhjkll... dora 4→ cut h

The three options which preserve the shape are ohl. Cutting o has no added benet.

Cutting l helps create ittsuu with g, but cutting h and aiming for junchan is best, with an

upgrade on any of jkl. While it returns to a bad shape and the yaku isn’t conrmed, with a

potential mangan it’s better than a ryanmen riichi only.

Ukase-uchi

In a shape with 2 groups, 2 joints and 2 pairs (14 tiles total), it’s usually best to break up a bad

shape joint, or to break up a pair if both joints are ryanmen. When deciding which pair to break

up, there is an opportunity for ukase-uchi, breaking one pair into a oating tile with upgrades into

better ryanmen than the hand already has. Otherwise, it’s best to emphasize safety.

Example eertysdghjccbn→ cut e

Can draw a sanmenchan or a ryanmen that conrms tanyao.

Example rtuioaadfgvvm,→ cut v

The oating v can be used to make sanshoku.

Example werttkkxcbnnm,→ cut t

Aiming to create a sanmenchan or non-overlapping ryanmen.

Example tyaahjccnm,555dora x → cut c

Can draw a dora ryanmen. We can also cut a if we suspect c would be too dangerous later

on.

Perfect iishanten

The basic idea when comparing dierent perfect iishanten is to compare the complex joints, focusing

on xing those that have good shape and xing shapes that create value. When there’s no big

dierence, we should emphasize an easy to win (on the outside) end wait.

Example wwessjklvvb

77→ cut w

1.5. IISHANTEN EFFICIENCY 95

While the qr is easier to win with than cn, the w is easier to call pon on than v.

While this is a dicult decision that is inuenced by the table situation, xing the good end wait is

usually better. Though this is in contradiction with the iishanten peak theory discussed earlier, the

improvement to the win rate from tenpai is more important here than the improvement to tenpai

rate.

The fact that makes iishanten peak theory work is that winning is possible from any player,

while entering closed tenpai requires a good tsumo. (There is also the advantage of riichi pressuring

opponents to fold) With closed perfect iishanten, the eective acceptance a is 20 tiles from 1-shanten

to tenpai, and b = 4×8 = 32 (slightly lower in practice) from tenpai to win. Since the expected

number of draws from 1-shanten to tenpai is 1/a and the expected number of discards from tenpai

to a win is 1/b, a marginal improvement in a is more valuable than one in b when a < b. However,

with a wider class of iishanten, a ≈ b, and iishanten peak theory no longer works as well. Similarly,

for an open perfect iishanten, a = 4×4+2×16 = 48 while b is still only 4×8 = 32, making iishanten

peak theory unreliable.

Extra tile versus perfect iishanten

Since perfect iishanten is wider, when there’s no specic reason it’s simply better. For the same

reason, a headless or sticky iishanten is better than a perfect one.

Don’t break perfect iishanten to keep a safe tile

The dierence in win rate is too large to justify a small reduction in deal in rate. This move

only makes sense in hands like errooxc4 where e is dora and eo have been cut

twice.

In the following examples we’ll mostly consider 2-shanten with 2 groups, 1 head, 1 complex

joint, 1 simple joint and 1 oating tile, with the option to take either an extra tile iishanten by

cutting a tile from the complex joint, a perfect iishanten by cutting the oating tile, or to stay in

2-shanten by doing something else. In all hands, assume East 1, 7th turn, dora 4. From the 11th

turn on, it’s usually better to take perfect iishanten, considering no-ten payments.

When the good shape is complex

Example weruggklzxcm,,→ cut u

Sacricing 4 tiles of immediate tiles acceptance for 8 tiles worth of ryanmen upgrade isn’t worth

it. However, if any of yuiowere dora, it’s good to cut ,, since there would be many upgrades

for better value. However, with no dora, we should also cut , if we want to avoid a riichi only

with a bad wait. When playing with aka dora, this is often the case.

Example weruaafhzxcm,,→ cut u

Here the inner kanchan has the same amount of ryanmen upgrades as the oating tile, so it’s

clearly better.

Example wertggklzxcm,,→ cut ,

Paragraph added by me to explain this fact more clearly.

96 CHAPTER 1. TILE EFFICIENCY

But here, there are 14 tiles for upgrade into ryanmen versus 4 tiles of immediate acceptance,

so , is good. If gg were hh, we would consider cutting l and returning to 2-shanten with

better shape.

Example werfggklzxcm,,→ cut ,

The oating tile next to the head is very strong, allowing us to make a good shape with any of

dgkl.

Example weruggllzxcm,,→ cut ,

The immediate loss in tile acceptance is only 2 tiles, while the amount of upgrades is very high.

This is a typical example of the theory that breaking up a 3rd pair is good.

Example wertfhllzxcm,,→ cut t or ,

Even on cutting t, there are still 8 tiles of upgrade to ryanmen. The dierence is subtle and

either of t and , can be good.

Example wertggklzxc,.. → cut l

Here, we think the return to 2-shanten with l is strong. This play is generally good in the

following kind of cases:

1. Advancing shanten makes a cheap hand with a bad wait likely

2. There are many tiles that revive the hand into a better 1-shanten (qerygm. in the

example)

3. It’s still the early game (6th turn or earlier)

When there is a possibility to upgrade into a better shape even when advancing, the criteria

become stricter. On the other hand, if the upgrades don’t just improve the wait, but also increase

value, they become looser. Furthermore, in the above example cutting , or . as a kind of

compromise is not a good plan, since a cheap hand with a bad wait remains likely if the manzu fail

to extend.

Example wertggkkzxc,.. → cut k

This hand is even narrower than the previous example, so declining 1-shanten is also good.

Example werggkklzxc,.. → cut l

While taking the 3 pair shape and hoping to draw f or h is good, cutting g is also playable.

Example qweyuuasdklxvn→ cut y

While this straightforward move into 1-shanten makes a cheap bad wait hand very likely, cutting

u into 2-shanten has very little promising upgrades and we would most likely end up with an

equally bad hand anyway.

1.5. IISHANTEN EFFICIENCY 97

Example ertysdfhhjbb,. → drop ,.

While this is big step backwards in shanten, conrming tanyao creates value and allows us to

call, compensating for the immediate loss in speed. (As mentioned in the subsection on tanyao.)

Example yuusddfkkn,555→ cut d

Since we can call pon, the loss from cutting u would be bigger. Since we have a yaku, we can

maximize immediate tile acceptance without waiting for an upgrade.

With a serial shape, a tile next to the head, a middle tile that creates value (or even a non-

middle tile if it’s guaranteed to produce a lot of value) we prefer an extra tile shape. If we want

to avoid a no dora no yaku bad shape riichi, we can keep even a regular middle tile. If the bad

joint has a tile acceptance overlap, breaking it up and returning to 2-shanten is often good (this

depends a lot on the turn, but in the early game we should return). If the bad shape joint is easy

to upgrade, perfect iishanten can be ne. With a 3 pair shape, we can take an extra tile even if it’s

a regular middle tile. But when calling is eective, the loss from not being able to call pon is big,

so we should only refuse perfect iishanten for a big increase in score.

If both our joints are bad, it’s often not a good plan to aim for an upgrade, since even with an

upgrade the end shape will be bad 2/3 of the time (because a good shape is easier to draw). Taking

the perfect iishanten to maximize tile acceptance can’t be helped.

When the bad shape is complex

Example wertasdhjllbm. → cut w

Since keeping the complex bad shape joint gives a bigger chance at a good end shape, we prefer

it over a connect 4 shape.

Example wertyuihjllbm. → cut .

But the connect 7 shape is so strong we can break up the ryankan (also seeing 567 sanshoku).

Example qqwerdfghcvnn,→ cut ,

Here, there are also upgrades to tanyao and 234 or 345 sanshoku.

Example qqrtsdfhjjbnnn→ cut j

The upgrades to a much more expensive (vbmm) hand are big. If n were dora, the value

is ne but we still want these upgrades for calling into open tanyao, so j is still good.

Example ertyaafgvvb... → cut v

Here, doubling the probability of sanshoku is big. While it also enables calling, a kata-agari is

not that great so if . were dora, we should cut either of ey.

Example tttyuaahhjzxcn→ cut h

Since the loss from cutting h is only 2 tiles, chasing even a far-o sanshoku is better. We can

also use the oating n to make pinfu. If we draw another manzu for a complex shape, we should

98 CHAPTER 1. TILE EFFICIENCY

cut n since the dierence in tile acceptance will ber very large.

With hands like these, holding an extra tile is only worth it when it has many upgrades for

better shape and also adds value. However, if an upgrade would double the value, we’ll often try

to preserve it, especially with hands that would only be worth 1300–2600 otherwise. If a hand is

already worth 3900 or more, it’s often best to maximize speed (a 3900 hand with riichi is actually

worth about 6300 on average). Only an upgrade to haneman would be worth sacricing a lot of

speed for, and this is quite rare.

When we have to cut dora

A closed hand with a oating dora that can cut it to make perfect iishanten or keep it for an extra

tile iishanten.

Middle tile dora In a hand worth 2600 without the dora it’s better than a serial shape, in a

hand worth 5200 it’s worse. In hands worth something in between, use discretion.

Dora is 2 or 8 If we can see 5200 and up without it, cutting it is good.

Terminal or guest wind dora Usually cut. If we have no other dora or yaku, we can keep it

depending on the point standing. If any win is ne, we cut it. If we want points, but a riichi only

hand is okay, we hold on to it as an extra tile. If we absolutely must win an expensive hand, we

should return to 2-shanten by breaking up a bad shape joint.

Yakuhai dora The risk of opponents calling it is big. We should especially avoid discarding it

if opponents have made threatening calls. If no such calls have been made, we should cut it with

≥2600 good shape or ≥5200 bad shape. With a cheaper hand, we should hold it as an extra tile

unless we don’t care about score. With two bad shapes, discarding a yakuhai dora for such a bad

hand is problematic, so we’ll often return to 2-shanten and aim for riichi with a dora pair or dora

tanki.

Dora at 2-shanten and before

As opposed to 1-shanten, keeping a oating dora while narrowing the shape doesn’t really decrease

win rate that much at ≥2-shanten. We’ll usually keep it until 1-shanten, and discard it only if

nothing sticks to it and another tile becomes more valuable. Conversely, when choosing between a

2-shanten with an extra tile dora which we would most likely discard anyway later, and a 2-shanten

which would lead to a clearly wider 1-shanten, we should prefer the latter. We mean a hand with

multiple complex joints, none of which we want to break up.

Example ttuuidffghcbnndora c → cut n

The loss from holding the dora close is small. (Since we have tanyao, we don’t really care about

losing out on chiitoi.)

Example eeerrtdgjzzcvmdora m → cut m

Cutting anything else is too much of a loss.

1.5. IISHANTEN EFFICIENCY 99

Cheap open hands

We mean hands that are worth 1000–2000 without holding on to the dora. Drawing a second dora

is a huge increase in score, but using only one is not as However, it’s easier to make dora tanki

since we can call. For this reason, it doesn’t matter as much as in closed hands how easy it is to

make a sequence with the dora (a dora tanki is easier to win with if the dora is dicult to use). In

these examples, we’ll take dora 4.

Example cvmm,aa4→ cut 4

Since the loss from not being able to call pon is big, we cut dora. If we need a big hand, we can

cut m.

Example xvmm,aa4→ cut 4

If we need a big hand, drop xv.

Example cvmm.aa4→ cut any of 4.a depending on situation

While dora is more important in open hands, the loss from not taking a perfect shape is also

bigger.

Understanding tenpai chance

A not very enlightening theoretical article reiterating many points discussed earlier, including

• Since the marginal value of tile acceptance is higher when it’s low, tile acceptance dierences

are less important with wider iishanten.

• This makes it more important to emphasize end shape and value.

• When comparing speed and value, speed is often more important.

• With a wider shape, playing by feel according to circumstances becomes more important

compared to theory.

• Not understanding the importance of tile acceptance and not understanding its relativity is

both bad. It’s important to balance these two. But chasing a balanced course for its own

sake without thinking of the optimum course is also foolish. Often, the optimal move will be

extreme while a compromise will be worse.

Headless iishanten

Can either have 1 joint that can complete into the head or into a group or 2 joints. The shape with

2 is much wider so we should almost always prefer it, unless a very good isolated tile is preferable.

Example qqqertfgzxcm.7dora 7 → cut .

With 2 aka, cut 7. With 1 aka, use discretion, depending on the point standing.

Example qqqertfgzxcnm7dora 7 → cut 7

100 CHAPTER 1. TILE EFFICIENCY

Since the dierence in tenpai chance is very big.

Example qqqertuihj45

5→ drop hj

The dierence in score is very large.

With a 1 joint shape, we should pay attention to keep the isolated tile which makes it easier to

create a head:

Example qqqifzxcbn,555→ cut i!

While the , in bn, is usually weaker than a normal i, this time drawing m creates a

good wait and drawing b also gives tenpai.

Similarly, even before 1-shanten, if we don’t have a head, number tiles in shapes like these

become slightly stronger.

Example rtuuioahklcvbm→ cut a

But if a more central tile exists that it’s easier to make ryanmen with, we should still prefer it.

Similarly, in a 1 joint headless iishanten, we should prefer a middle tile to a safe tile.

Example errtasdghcvbm4→ cut 4

Extra tile versus headless

With a shape like qqwe or qwwe, xing a group can give a headless shape, or xing into

the head can give an extra tile shape. If we have a closed set or a serial shape, we should

prefer the headless shape. This is because they increase the tile acceptance for ending up with

a good wait.

Example qwwesfghjzzznm→ cut w

Example qwwefghjkzxcnm→ cut w

Example qwwefgghhzxcnm→ cut w

Example qwwehjjklzxcnm→ cut w

Example qwwehhjkkzxcnm→ cut w

But with a serial shape like qwwee, it’s actually harder than usual to make a head, so

it’s better to x the head.

Example qwwejjkklzxcnm→ cut e

There is no benet to drawing r so we keep q for safety.

Example erassddghjkcvb→ cut d

Aiming for tanyao and/or 345 sanshoku. The assdghjk shape completes with any

of fgl.

1.5. IISHANTEN EFFICIENCY 101

Example uuiioasdghjkcv→ drop cv

Hoping to extend pinzu into a better shape.

Without a serial shape

With all ryanmen and no serial shape, the headless shape has a wider tile acceptance, but the extra

tile shape has a wider tile acceptance towards a good wait. This is a famous shape where it’s easy

to hesitate.

Example qwweasdhjcvbm,→ cut w or e?

We should decide depending on the discards, choosing the option with more tiles left. If the

hand is too cheap to attack with a bad shape, taking the guaranteed good shape iishanten is

recommended. Keeping an extra tile can also be advantageous for other reasons (see above).

If any of hjm, is dora, dora tanki riichi is not bad, so the headless shape is stronger. If

we have a dora in one of the sequences, we can draw another one and turn it into the head if we

are in the headless shape, so it also becomes slightly stronger.

Example qwweaassddhjm,→ cut w

With a yaku, tanki riichi is ne.

But if we can make yaku, we should of course prefer that shape:

Example qwwesdfhjcvbm,→ cut q

Example qwwesdjklzxcbn→ cut w

With at least one bad shape joint, taking the headless shape is better. If we have a werr

aryanmen shape, we can break up the weak joint to make an extra tile iishanten, hoping to draw

a better shape headless iishanten later. The same can be done with a dffg nakabukure if

the turn is still early. When not in 1-shanten, xing a qwwe shape into a group is usually

good since it’s easy to draw a head elsewhere. The exception is usually only when we would have

to discard dora.

When a shape that would allow us to exploit a headless shape eectively (closed set, serial shape

etc.) exists or upgrading into a headless iishanten would in another way be good, it’s often good to

keep an extra tile next to the head in an extra tile iishanten. Otherwise, it’s better to keep a safe

tile.

Serial shape versus score

Example qweeyudfghjvbn→ cut q

Acceptance is 19/19 with q versus 26/37 with e. But cutting q conrms tanyao, doubling

score, so it’s better despite being narrower. If we draw r afterwards, we should cut e and go

headless. If the pinzu shape was fghjk, the problem is more dicult. We should still cut

q in the early game or when we need an expensive hand.

102 CHAPTER 1. TILE EFFICIENCY

Example qweedfghjvbnnm→ cut e

Acceptance is 18/18 with qversus 42/42 (!) with e. With a double serial shape, the dierence

in tile acceptance is overwhelming. Unless we absolutely need at least 3900 to increase placement,

e is better.

Open hands

Example qweeyufgvbn77

→ cut q

Since calling with a headless shape always results in a tanki, taking the extra tile shape is better

here.

Example wryssfgxcv

n,scomes → don’t call pon

Example wryssfhxcv

n,scomes → calling pon is playable

Calling pon will increase the tile acceptance in both examples. However, we also give up a

tsumo and make the hand shorter, making it harder to defend. Since in the rst example calling

makes it slightly harder to end up with a good shape, it’s better to not call, while we should at

least consider calling in the second one. If xcv were instead bbb, we should call and cut

w for a clearly better anko headless shape, also being able to stick a ryanmen to b.

Perfect versus headless

We’ll center on hands with a many-sided wait. While the headless shape will be wider, the perfect

shape will have the advantage of leaving a many-sided wait. As always, assume 7th turn in East 1

and dora 4.

Example wwweuiojkxc333→ cut e

If we want to keep the 3-sided wait, we’d have to break up either of the ryanmen. We call the

shape on cutting c the xed form, on cutting e the owing form. The tile acceptance for the

owing form is 28, and 22 for the xed form, of which 8 tiles give a 3-sided wait. However, the

dierence in win rate between a 3-sided wait and a 2-sided wait is smaller than between a 2-sided

and a 1-sided wait, so we should simply prefer the tile acceptance.

Example wwweertjkxc333→ cut e

Here we have a 4-sided wait, with 24 tiles accepted in the xed form, of which 8 give a 4-sided

end wait. While this is a lot closer than the previous example, a 1/3 chance at getting an improved

end wait is not that great, so the owing form is slightly better.

Example wwwertyjkxc333→ drop a ryanmen

Here, the two forms have the same acceptance, but the xed form oers an excellent 5-sided

wait. We should look at the discards and drop the worse ryanmen.

Example wwweuiojlxc333→ cut l

1.5. IISHANTEN EFFICIENCY 103

Here, taking the owing form leaves the possibility of a bad wait (10 good/24 total). But cutting

l (we can draw h and cut e then) gives a tile acceptance of 22/22, which is a lot better.

Example wwweuiojlxv333→ cut e

The owing form has 0/20 acceptance, the xed form (we cut from the outside, hoping to draw

a ryanmen upgrade, and we prefer to preserve the jl as it has the same amount of upgrades

but is easier to win with) has 4/18 acceptance, with the 4 being for a 3-sided wait. Here also,

emphasizing the end shape is better.

Example wwwruiojlxc333→ early game cut l, else r

The owing form has 10/24 acceptance, the xed form 18/18. This is a dicult problem. This

dierence in total tile acceptance is the dierence between an average of 5.5 and 7.5 turns to tenpai.

Especially from the mid game on, this dierence has a big eect, so we should cut l early and r

later.

Example wwweuiojlxcm,. → cut e

Here we no longer have a closed set, but cutting e anticipates sanshoku. Since the possible

increase in score is very large and the dierence in speed is smaller than between a ryanmen and a

kanchan, it’s better to aim big. When choosing between the xed form and owing form, choose

the one with acceptance towards a big hand. However, if we already had dora 2, the added

value from sanshoku becomes inecient. With a conrmed mangan, it’s better to emphasize

speed.

Example ertteuiojlxc333→ cut l

Similar to the example with wwwrjl above, but the manzu shape is easier to extend.

Example weryisdfhkbnm,dora e → cut b

Since breaking a bad shape joint when we have two will often lead to bad wait tenpai anyway,

and the b, wait for 6 tiles is not great, it makes no sense to keep the serial shape. We maximize

tile acceptance and can accept another dora to make 678 sanshoku.

In short, if the owing form guarantees a good wait, we should take it unless we miss out on a

very good many-sided wait. Since a 2-sided wait is already good enough from a win rate perspective,

it’s often better to emphasize the speed of getting to tenpai.

However, there is often not a big dierence, so if one option is clearly more expensive, we should

prefer that. We can also use information from discards to emphasize the suits that look easier to

draw or come out from opponents.

Example eeertyufgjklvb→ cut u

If one of the joints in the owing form would be bad shape, we should prefer the xed form by

a wide margin. However, there are some cases where we can take the owing form if the bad shape

joint creates value or is particularly easy to upgrade. On a late turn, we should also emphasize

pure tile acceptance to get into tenpai as fast as possible.

This example feels out of place, yeah

104 CHAPTER 1. TILE EFFICIENCY

Problem East 1, 6th turn, dora x Source: Japan Mahjong Organization

weerrttydgcvbb

Tsumogiri b, keeping the widest shape with tanyao and sanshoku. When drawing r or u,

enter provisional tenpai for tanyao and try to improve the wait. When drawing q, ignore tenpai

and tsumogiri.

With two bad shape joints, the reasoning is similar to two good shape joints, taking the xed

form only when the many-sided wait is strong. The same ideas about value and the turn as above

apply.

Sticky iishanten

This shape is the widest of them all, and especially in the late game where raw tile acceptance into

tenpai for no-ten payments is important, we’ll rarely prefe another shape.

When comparing isolated tiles, we should be aware that unlike at pre-1-shanten, we don’t care

about making ryankan, so we should not cut c from euc, but one of the other two since they

overlap.

Example werudfghllzxcm→ cut u

While usually being able to draw t for an easy to upgrade kanchan is an advantage, here it’s

actually an overlap since we can use it to widen the shape even if we discard u.

Example ertydffxcvbnm,→ cut y

While d is widest at 35 good/60 total, y gives 39/46. While it’s only 4 tiles of good wait

versus 14 tiles of acceptance, if the hand is this wide, it’s better to emphasize end shape. We can

also see 345 sanshoku.

Example wertyuugghvvbn→ cut v

While h gives an acceptance of 45, only 20 of those give a good shape. v gives 40 tiles

of which 26 good shape, so it’s even better than in the previous example. We can also see 456

sanshoku. While gcan give both 456 and 567, it reduces tile acceptance to 23/23 and is therefore

not recommended.

Example wertffgxcvb666→ cut g

Cutting ggives tile acceptance 30/50, cutting wgives 30/41 which is strictly worse. The cause

is that unlike the examples above, this hand doesn’t have a 3-sided shape like wertyuu

or wertyui. While cutting w appears to give 345 sanshoku, we’d need to get either x

or g together with d, which is rare (7/41).

Tsuchida Kōshō-chaired organization that disbanded in 2010

1.5. IISHANTEN EFFICIENCY 105

Example errtyuifggbbnm→ cut f

This hand also doesn’t have a 3-sided shape. The acceptance with f is 28/41, with b 25/36.

f is clearly better. In general, a nakabukure is dicult to turn into the head, so xing a head is

often good.

Example rrttyyuffghbnn→ cut u

u gives 23/37, b gives 27/43, f gives 26/38. All of these are super wide and likely to give a

good wait, but the u which conrms iipeikou and enables sanshoku and ryanpeikou is a lot better

for score.

Example weefjklbbnm555→ cut f

Better because it’s isolated (23/33 versus 19/41 for w).

To recap what we have seen so far:

• 3-sided serial shape + 2-sided serial shape + ryanmen pair ⇒ cut from the 2-sided serial shape

• serial shape + isolated tile + ryanmen pair ⇒ cut isolated tile

• 2×serial shape + ryanmen pair ⇒ x pair, even more eective with a kanchan or penchan

pair

The relative strength of serial shapes goes as xcvbnm, > cvbn > cvvb

cvbb

Example errdfghccvbnm,→ cut c

e gives 25/53, c gives 30/35, g gives 25/40 (exception to the above, since the 3-sided

aryanmen is weaker than a regular 3-sided serial shape). Emphasizing a good wait is better.

Example errddfgccvb555→ cut d or c

e gives 20/42, the other two give 23/37. The acceptance into sanshoku is 0 versus 8, into

iipeikou 12 versus 6, so d or c is better. If sanshoku wasn’t possible, e would be better. (A

bad shape conrmed iipekou 5200 is better than good shape 2600.)

Problem dora r (Source: Japan Pro Mahjong Federation)

rtttyufghjbnnm

Answer: n is the widest, making ryanmen tenpai even with fj.

Problem eerpyuvbbnm,,.

Answer: e. While , creates a sticky shape which is wider, it makes it impossible to achieve

tanyao, so w is better. While . is playable, the lost acceptance into riichi with bm, is more

106 CHAPTER 1. TILE EFFICIENCY

valuable than being able to call pon for tanyao. This a rare exception where declining a very wide

sticky shape is better to increase the score, compensating with a strong sanmenchan + complex

ryanmen pair.

Perfect versus sticky

We usually prefer the much wider sticky shape, but cases where the perfect shape is preferable do

exist. This is often when a 3-sided wait with a closed set (weee) and a ryanmen pair are in

the hand. If we have a head elsewhere, the isolated tile next to the closed set becomes weak. Often,

while the sticky shape will have more total acceptance, the perfect shape will have more acceptance

towards a good wait.

Example qweudddfjklnmm→ cut u

Example qwedddfjkkbnm,→ cut f

The serial shape in souzu makes it easy to get a good shape with a sticky shape. But cutting

k has the advantage of being able to make an extremely wide headless shape when we draw the

backre of vnm..

Example eersdfhjjjxcvb→ cut e until the mid game

While cutting h is the widest overall, cutting e makes it easier to get tanyao pinfu for a

higher score.

Example errtyyydddfnnm→ cut n

With 2 many-sided waits, preserving them is clearly good (28/28 compared to 22/33 with r).

We usually won’t get sanankou so we ignore it.

Returning to 2-shanten

When to return

Returning to 2-shanten is limited to extra tile or perfect iishanten with bad shapes, or chiitoi

iishanten. (Except for special situations where an expensive hand is required at any cost.) We

should never return to 2-shanten after the 12th turn unless we’re folding, because we can get no-ten

payments even with a very weak hand, unless the point dierences are very large and we don’t care

about no-ten payments.

Patterns of returning

Assume East 1, West seat, dora unrelated to the hand.

Protip: with problems like these, http://kobalab.net/majiang/dapai.html is a great help (it doesn’t do open

hands, but can be set to aka nashi). Often the expected value dierence between two good moves will be very small.

Try it on some of the examples in this chapter and see where it disagrees with the author!

bfor sanshoku is also not bad

1.5. IISHANTEN EFFICIENCY 107

Excellent oating tile (connect 4, nakabukure)×2

Example wertaadfjlnmm,→ cut l

Excellent oating tile + value-creating oating tile

Example qqyasdghjkzxnm→ drop zx for ittsuu/sanshoku

Example eriossfghhvbnm→ cut o for tanyao

Value-creating oating tile ×2

Example uoffghjkl33445→ drop uo

Example qqweuolmmm,.44→ cut m

Since creating a yaku enables us to call, this compensates for the loss in speed.

Extremely bad shape (overlap, less than 2 winning tiles) or conrmed worthless bad

shape hand

Example eeryoodhjjkbnm→ cut y

Example qeiossfghhvbnm→ cut o

Doesn’t give tanyao immediately, but a worthless (no yaku or dora with bad shape) 1-shanten

is not desirable anyway.

Example werioaaghhvbnm(2 u left) → cut o

If 4 u are left, cut h.

Going from chiitoitsu 1-shanten to a wide or freely callable normal 2-shanten

Calling from chiitoitsu 1-shanten into toitoi 2-shanten, if chiitoi would have no yaku

or dora and toitoi would be conrmed mangan Especially with honitsu + yakuhai where

the hand would be expensive even with a sequence.

These are roughly the main cases, but in general we can decline a weak 1-shanten if we have

suciently good oating tiles. While we should be more prepared to stay in 2-shanten in the very

early game, if we have only 2 or 3 upgrades, we should take a weak 1-shanten even extremely early.

The reason is that we’ll have only 18 draws at best, and waiting for these upgrades will often take

much too long (11 turns on average for 3 kinds, 17 for 2 kinds).

108 CHAPTER 1. TILE EFFICIENCY

Easy to upgrade 2-shanten versus expensive 1-shanten

Even if the 2-shanten would be easy to upgrade, if the 1-shanten has a possibility of making a big

dick hand, the loss from returning is bigger. This is often a dicult problem. In the early game,

we’ll usually return, and preserve 1-shanten from the 7th turn on. We should also consider the

speed of the opponents.

Problem East 1, East seat, 7th turn, dora h

rrtsffghjvbnmh

Answer: m. While cutting s into 2-shanten has many upgrades, the loss from missing out on

dgk is very big. Rather than cutting t to keep the serial shape, we retain the possibility of

sanshoku, a perfect iishanten with sj and a sticky iishanten with ey.

This concludes the chapter on tile eciency. With the contents so far, we can usually choose

the right discard in an equal point situation when no opponents are attacking. From here, we’ll

take up other skills and some practice problems on tile eciency.

Chapter 2

Push-fold judgment

2.1 Dealing with non-tenpai opponents

Choking

So far, we’ve discussed ways to determine the best move without considering opponents. However, if

an opponent is in tenpai, we can’t just discard tiles we don’t need. If we want to defend rather than

attack, we should deal the tile least likely to be useful to opponents (folding). Deciding whether to

push or fold will be the subject of this chapter.

However, we can also say something about situations where no opponent is in tenpai, but we

still should be careful. One of these is choking (shibori). We mean purposely not discarding a tile

which we don’t actually need to prevent opponents from calling it.

When to choke

Cases where we should choke in the strict sense (merely to prevent opponents from calling, not to

not deal in or to use the tile ourselves) don’t really exist. That’s because we should discard tiles

that are useful to opponents early. The probability of dangerous tiles being called or getting

ronned increases continuously. It’s better to get rid of the tile early and increase our own chance

of winning.

Example East round, North seat, 6th turn, dora 1

erdf;hjvbn,,13→ cut 1

This hand is mentanpin aka closed, so fairly sure of mangan. Even if we draw a second 1, we

still only get mangan. So we cut get rid of it early. If we didn’t have tanyao, we’d cut 3 instead,

since drawing a second 1 would double our score. If 1 weren’t dora, it’s useless and we should

get rid of it.

Even if a tile looks very likely to be called by an opponent going for a particular yaku, it’s often

not worth it to choke, since this will slow down only ourselves and one opponent, beneting the

other two opponents. We should only choke if we don’t care about helping the other two

opponents, most often in the following kind of cases:

109

110 CHAPTER 2. PUSH-FOLD JUDGMENT

• An opponent has made a call for a clearly big dick hand (for example, choke yakuhai when

someone pons terminal dora)

• With a big lead, we only need to prevent dealer or 2nd place from winning

If no one is in tenpai, it’s generally not necessary to care about opponents.

Dealing with getting choked

It can be good to refrain from calling if opponents would try to choke us in the following kind of

cases:

• Calling would lower the score

• We don’t have enough blocks

• We intend to fold soon

• We have an especially obvious yakuhai atozuke

Discarding good tiles rst

If we have multiple useless tiles, it can be good to discard the ones most dangerous towards oppo-

nents rst.

Early game

Example West seat, 1st turn, dora a

wtasshhm,22456→ cut w

In this hand, we cut w over 4, but instead of keeping 4 as a safe tile because the hand is

bad, it’s more to create yaku (honitsu, chiitoi) more easily, making it a more useful tile. If we make

honitsu, we can keep it over the o suit tiles as a safe tile while staying ecient. With yaku that

restrict the usable tiles, keeping an honor over the disallowed tiles is common, allowing us to fold

easily against a riichi (we can often fold successfully if we have 3 safe tiles). However, if riichi is the

mainline, we should do everything to maximize win rate, so it’s not needed to keep safe tiles in the

early game.

1-shanten

Example West seat, mid game

wweuidfgjklzz4→ cut 4

We should almost always prefer an eective tile over a safe tile.

Example West seat, mid game

wetydfgjklzzn4→ cut 4

2.2. FOLDING TECHNIQUES 111

Even the upgrades into a wider are more valuable than a safe tile. In the late game, the two are

about equal.

When to keep safe tiles

We should discard tiles that are guaranteed safe but especially likely to become dangerous in the

future. For example, if an opponent is going for honitsu in souzu and n passes from a dierent

opponent, we should discard it even if it’s useful, especially since a ryanmen in souzu would be

dicult to win with.

If we’re still in 2-shanten or worse in the mid-late game, have low tile acceptance and can’t call,

our chance of winning is very low. Holding on to safe tiles with a hand like that is wise. However,

it’s better to hold on to several than to just one, and often it will be good to just fold completely.

The same is true about defending against damaten. If the hand is narrow and 2-shanten or

worse in the mid-late game, we should start being cautious about damaten.

Early game defense

In the early game when no one is in tenpai, it’s not really necessary to play defensively.

Even when leading in all last, the dealer can renchan indenitely and close any gap, so we should

try to end the match if we’re not the dealer.

Exception: player 50000, dealer opponent 25000, opponent 24000, opponent 1000. In this case,

even a big dick tsumo from the dealer won’t endanger us, so we can fold immediately to prevent

dealing in.

Exception: dealer opponent 55000, player 35000, opponent 8000, opponent 7000. Unless we can

realistically make a big dick hand, we can fold to avoid dealing in and secure 2nd.

If we’re dealer and have a big lead in all last, we only need to endure one more round. But

because dealer pays double on tsumo

, if an opponent could overturn us with mangan direct hit,

he could also haneman tsumo, and similarly for haneman direct hit and baiman tsumo. So instead

of going for a direct hit with dama, he could just riichi and tsumo, or instead of going for direct

hit with riichi, he could riichi tsumo ura/ippatsu.

That’s why even with a big lead, giving up on winning and stockpiling safe tiles is not a good

plan. Furthermore, depending on the point system, winning a hand can give a better point outcome

after the match. Of course, from the mid game on, if our hand looks hopeless we can fold against

dama, or fold against a riichi or call even from tenpai.

2.2 Folding techniques

Betaori

We’ll now discuss situations where an opponent is in tenpai. When we’re not in tenpai ourselves,

we usually should betaori. What this means is to give up on winning and discard tiles that are

least likely to be the opponent’s winning tiles. Since rounds where we won’t win outnumber rounds

where we’ll win greatly, learning betaori has a great impact on results.

This entire chapter is heavily based on Scientic Mahjong by Totsugeki Tōhoku and the author often gives no

explanation for why a move is good, referring to that book instead.

This is called oya-kaburi, “dealer hat”.

112 CHAPTER 2. PUSH-FOLD JUDGMENT

Danger level of tiles

Please memorize this diagram.

Diagram of tile danger level

rank type deal-in rate

S genbutsu 0%

A+ tanki honor 0.9%

A none

B suji 1 and 9 2.9%

C non-tanki honor 3.4%

D suji 2 and 8 4.8%

suji 3 and 7 5.5%

E no suji 1 and 9 6.3%

half suji 4, 5 and 6 7.0%

no suji 2 and 8 7.0%

no suji 3 and 7 7.1%

F no suji 4, 5 and 6 12.3%

Genbutsu are tiles discarded by the opponent or any tiles discarded after a riichi. Tanki honors

are those that can only be targeted by a single wait, so those of which 3 are visible. Double suji 4,

5 and 6 (r with both q and u safe) are comparable to suji 2 and 8, so rank D.

Factors that aect danger level

Kabe (no chance) is equally safe or safer than suji If all 4 e are visible, w can’t be

targeted with a kanchan, which it can if it’s suji. No chance tiles become safer the more of them

are visible (less chance of shanpon/tanki, this is also true for suji and honors). If all f are visible

and l is safe, h is also no-chance. Similarly if all f and k are visible. This is especially easy

to overlook.

One chance is safer than no suji but less safe than suji

Tiles outside of early discards are 1 rank safer About 60% as dangerous, even safer against

open hands. (Cutting e from wee early with an open hand is rare.) The earlier the discard,

the safer. The same thing is true of suji, so the suji of the riichi tile (ryankan suji trap) is slightly

more dangerous. Between a tile outside of an early discard and a suji of the riichi tile, we should

pick the former. (Note: since in kuitan nashi calls are less common, this technique becomes slightly

less reliable.)

Ura suji, aida yon ken No need to consider this.

Neighbors of the riichi tile (matagi suji) Sometimes more dangerous, but usually no need

to consider this.

2.2. FOLDING TECHNIQUES 113

Anko suji (The suji of a closed set in our own hand) Slightly more dangerous, but not by an

entire rank.

Dora and neighbors The dora itself is 1 or 2 ranks more dangerous (120–130% more dangerous

for tanyao tiles, 170% for terminals). Tiles next to the dora are about 1 rank more dangerous

(110%).

Suspicious discards (at least 3 dierent 456 tiles) Not a factor of itself.

A player who discarded an aka usually doesn’t have a regular 5 of the same suit (when

playing with 3) When ; has been discarded, dj are rank B (can only be targeted by penchan),

since from dd; cutting d is standard, and an aka tanki is preferable to a dj tanki. If ;

has been discarded and a is safe, f becomes much safer and similarly for ;, l and h. If an

opponent discarded /, the suji x, are safer than regular suji, since he usually won’t have a

ryankan suji trap (but aka traps do exist with some hands).

Discard order for equally safe tiles

When folding completely, consider the possibility of other opponents entering tenpai and keep

common safe tiles, discarding the rest rst. Conversely, when still considering our own

tenpai, we should fold with the tiles that are least useful to our own hand.

Among equally safe tiles, we usually want to discard the tile that opponents can call. This

is because opponents counterattacking into each other decreases the chance of having to pay for a

tsumo. However, if a draw looks likely and the open hand would be so expensive as to endanger

our position, it’s better to not let opponents call.

Eectiveness of wait reading

The theories of ura suji

and matagi suji have mostly been disproven.

Why ura suji is wrong

For example, the ura suji of w are e and y. The old theory is that from a wrt shape,

players would cut w early, so an early discard of w would make its ura suji more dangerous.

However, discarding a regular w also usually happens early, and we can’t distinguish between the

two. Furthermore, discarding w early makes e slightly safer by making a kanchan or penchan

less likely. In general, this kind of ura suji has a negligible impact.

However what about the ura suji of b, which are zvn.? This is because a b is a very

good tile, and discarding it early is rare. It only makes sense in a xcb or bm, shape, so an

裏スジ (n): frenulum of prepuce of penis

114 CHAPTER 2. PUSH-FOLD JUDGMENT

early b often indicates those shapes. While vn are already very dangerous anyway, if a b has

been discarded early, z. do become appreciably more dangerous. Similarly, a no suji , with

an early v discard also becomes more dangerous, but not enough to make it F rank. Since we’ll

rarely need to fold with tiles that dangerous anyway, this doesn’t really matter in practice.

Why matagi suji is wrong

The idea of matagi suji is that when the riichi tile is t, ey and ru become more dangerous

since the t will often come from a ryanmen pair. However, since a t is so strong, there are

an innumerable amount of other shapes where a player might want to keep it until tenpai. Also,

making no suji middle tiles more dangerous is rarely relevant.

However, if the riichi tile is k, hl do become more dangerous. This is because an isolated

k is rarely better than a middle tile, so shapes likes jkk and kll are especially common.

(This also is less than a 1 rank dierence.)

Principles of discard reading

Discard reading is the most useful when an opponent has discarded a very useful tile early, or a

usually not very useful tile late, since these allow us to infer the existence of particular shapes in his

hand. Similarly, an opponent who has discarded joints or pairs can usually be read to be holding

better ones. Conversely, an opponent who has only discarded honors before riichi gives us no hints

at all. Reading is only helpful when using the basic diagram from the previous subsection and

counting the number of remaining suji, and applying especially strong hints from discards to the

remaining tiles to pinpoint particularly dangerous tiles. However, when folding completely we’ll

often have many safe tiles, while with a strong hand it’s usually good to push even slightly more

dangerous than normal tiles. Discard reading is usually limited to situations where the decision

between attack and defense is hard, or when we want to defend but have no safe tiles.

Coping with having no safe tiles

What should we do if an opponent has entered tenpai and our hand is unfavorable for counter-

attacking, but we have no safe tiles? By this, we mean that all our tiles are rank E or F on the

diagram shown above.

For the time being, we should attack, and fold when we get more safe tiles. If we

have a bad shape iishanten with no safe tiles in the early game (because we didn’t keep safe tiles

to maximize speed), there will be many suji that have not yet passed, so risking one of them will

not be that dangerous.

However, if we have a 2-shanten or worse in the mid-late game and run out of safe tiles, our

chances of winning would be hopeless even with an all-out attack. In this case, it’s wiser to try to

minimize deal-in rate and deal-in score even by a little. The following techniques can be helpful:

2.3. DEALING WITH RIICHI 115

Cutting from a set or pair, or q→r when u is safe, forcing through a dangerous

tile to create safe tiles for later Safer than cutting multiple dierent no suji tiles. Of course,

between multiple no suji tiles, we should prefer to cut the safest ones (terminals) rst.

Cutting tiles that would lower the deal-in score Especially terminals that would deny

tanyao, and avoiding tiles close to the dora. But because a riichi can have ippatsu and ura dora to

inate its value unpredictably, this method isn’t very reliable. It’s more useful against open hands.

Cutting tiles conjectured to be safe from discard reading Or rather, not cutting tiles

that look especially dangerous. We’ll explain this in more detail in the chapter on discard reading.

This technique is also more useful against open hands because a riichi will often have irime. (For

example, if an opponent declared riichi with a i, he often had a uii in hand, making a yo

more dangerous. But half of the time, he drew that wait rst and is now waiting on his other wait,

which would make yo irime, “entering piece”.)

2.3 Dealing with riichi

The average score of riichi

Whether to attack or to defend depends heavily on our own hand. We’ll rst consider the case

where we’re in tenpai, since otherwise we have to make estimate our chance of reaching tenpai,

complicating the calculations. From 2-shanten or worse, to fold is usually the right answer, but

from 1-shanten the decision will be especially hard.

Since with open hands, judging whether the opponent is in tenpai is not straightforward, but

we can infer a lot about his hand from discards and calls, we’ll consider the case where an opponent

has declared riichi rst and we’re also in tenpai.

It has been shown from statistical analyses of online play that with 3 aka dora in play, the

average score of a winning riichi is 7000 for non-dealer, 9800 for dealer. (Versus 6000 and

8400 with no aka.) While we’ll continue the rest of the chapter assuming the former, the dierence

is not very big and in aka nashi we should play only slightly more aggressively with the same value.

When in tenpai

Assumptions

• Ippatsu, ura ari, 3 aka, shuugi

nashi.

• If our hand is closed, we’ll declare oikake riichi. (This is almost always advantageous unless

our wait is on a safe tile. We’ll discuss examples of oikake dama later.)

• We can safely avoid dealing in if we fold. We have enough (3 in the mid game, 2 in the late

game) safe tiles (B rank or up).

The retarded mahjong parlor rule where a set price (typically ¥1000) is added to the score for each aka dora

(often also for ippatsu, ura dora and rinshan)

116 CHAPTER 2. PUSH-FOLD JUDGMENT

• The danger level of F rank tiles in the early game (4th turn), E rank tiles in the mid game

(7th turn), D rank tiles in the late game (12th turn) is roughly 5%.

• The danger level of F rank tiles in the mid game and E rank tiles in the late game is roughly

10%. (The more safe tiles there are, the less safe the rest becomes.)

• A lot of data is taken from Scientic Mahjong, so it’s for an average situation. If we have spe-

cial knowledge about the opponent’s style or the table situation, we can change our assessment

appropriately.

• We don’t really care about no-ten payments our riichi sticks. If no-ten payments are impor-

tant, we should be slightly more aggressive in the late game.

We’re dealer

• With a good shape

– We should always attack. (Especially when considering renchan and no-ten payments.)

– Only fold when we’re okay with losing our dealership and especially want to avoid dealing

in, or when we would have to push an extremely dangerous tile.

• With a bad shape

– In the late game (12th turn)

∗ Cutting a tile with 10% danger level (E rank), with at least 3900 riichi.

∗ Cutting a tile with 5% danger level (D rank), with at least 2900 open.

– In the mid game (7th turn)

∗ Cutting a tile with 10% danger level (F rank), with at least 3900 open.

∗ Cutting a tile with 5% danger level (E rank), with anything but a 1 han hand.

– In the early game (4th turn), we should push everything. (Discarding a 10% danger level

tile with a 1 han hand is slightly disadvantageous, but tiles are very rarely so dangerous

this early.)

Non-dealer versus non-dealer

• With a good shape

– In the late game (12th turn)

∗ Cutting a tile with 10% danger level (E rank), with at least 2000 riichi or 2600 open

(riichi only and 2000 open are borderline).

∗ Cutting a tile with 5% danger level (D rank), anything except 1 han open.

– In the mid game (7th turn)

∗ Cutting a tile with 10% danger level (F rank), with at least 2000 open (riichi only

is borderline).

∗ Cutting a tile with 5% danger level (E rank), usually attack with anything (1 han

open is borderline).

2.3. DEALING WITH RIICHI 117

– In the early game (4th turn), we should push everything.

• With a bad shape

– In the late game (12th turn)

∗ Cutting a tile with 10% danger level (E rank), with at least 5200 riichi or 5200 open.

∗ Cutting a tile with 5% danger level (D rank), with at least 3900 open (2600 riichi is

borderline).

– In the mid game (7th turn)

∗ Cutting a tile with 10% danger level (F rank), 3900 open is borderline.

∗ Cutting a tile with 5% danger level (E rank), with at least 2600 riichi.

– In the early game (4th turn), same as in the mid game. Attacking with a cheap hand

with no safe tiles and folding once we get some is an eective strategy.

Against the dealer

• With a good shape

– In the late game (12th turn)

∗ Cutting a tile with 10% danger level (E rank), with at least 2600 riichi or 3900 open.

∗ Cutting a tile with 5% danger level (D rank), with at least 2000 riichi or 2600 open

(2000 open is borderline).

– In the mid game (7th turn)

∗ Cutting a tile with 10% danger level (F rank), 2000 riichi is borderline.

∗ Cutting a tile with 5% danger level (E rank), with at least 2000 open (riichi only is

borderline).

– In the early game (4th turn), same as in the mid game, with the same caveat about

folding later being a good strategy.

• With a bad shape

– In the late game (12th turn)

∗ Cutting a tile with 10% danger level (E rank), with at least mangan.

∗ Cutting a tile with 5% danger level (D rank), 5200 riichi or open is borderline.

– In the mid game (7th turn)

∗ Cutting a tile with 10% danger level (F rank), 5200 riichi or open is borderline.

∗ Cutting a tile with 5% danger level (E rank), with at least 5200, 3900 open or 2600

riichi is borderline.

– In the early game (4th turn), same as above.

Recap

What we understand from this is that counterattacking is much better with a good shape,

and the requirements for a bad shape are quite strict, especially when facing the dealer. When

choosing which tile to call riichi with, we should also heavily emphasize a good wait. If calling riichi

with a head start, 5200 bad shape is better than 2000 good shape, but when chasing, the latter is

better: we should pick the good shape even if it halves value.

118 CHAPTER 2. PUSH-FOLD JUDGMENT

Choosing the wait against riichi

Cheap good wait or expensive bad wait

We should pick the the good shape even if it halves value. (Even riichi pinfu > riichi sanshoku,

since their average winning scores are about 3500 and 7100.) If the dierence is bigger than a factor

of 2, the bad wait is usually better.

Cut a dangerous tile into a good wait (or safe tile wait) or a safe tile into a bad wait

Unless a draw is close or we intend to go dama and fold immediately, we should cut the dangerous

tile to increase our win rate. This will lower our deal-in rate in the end.

Cut a dangerous tile for an expensive hand or a safe tile for a cheap hand

Unless the dangerous tile is especially dangerous, it’s a loss to cut a safe tile to halve our score.

Should we go dama if our wait is on a safe tile?

With a good shape, if we have at least mangan. Ryanmen dama 3900 or 5200 is riichi, 6400

or 7700 is dama. However, if we can’t expect going dama to increase our win rate (opponents aren’t

folding anyway or we are clearly pushing many dangerous tiles so they get wary and start defending

against us), we should riichi anyway. Of course, if our hand is too cheap to chase, we should fold.

With a bad shape, we should usually dama. But similarly, if we can’t expect a higher win

rate, we should riichi or fold. Since determining how much of a win rate increase we can expect

from going dama depends heavily on opponents, we’ll need to apply discard reading skills.

Dealing with 2 riichi

Good wait, cutting a common safe tile (in this subsection, we consider an honor shanpon

a bad wait, as we care more about tiles left in the wall than tricking opponents) On any turn,

win rate is about double deal-in rate, and ippatsu is common. Unless we especially want to avoid

dealing in, we should chase even with riichi only.

Bad wait, cutting a common safe tile Win rate is roughly similar to deal-in rate, so we should

attack if our hand has similar value to the opponents. If our hand is too cheap, we should take

dama and fold if we draw a safe tile (or call riichi if we can upgrade the wait or value). If we have

no other common safe tiles, we should chase. (But because a riichi duel usually ends quickly and

safe tiles are added more quickly too, we can usually fold successfully with only 2 safe tiles.)

Good wait, cutting a tile dangerous (10% danger level) against both Deal-in rate is

slightly higher than win rate, so we should attack only with a mangan. The same thing as above

holds about the number of safe tiles.

Bad wait, cutting a tile dangerous against both Deal-in rate is more than double win rate

(even more in the late game). We should only attack with a huge dick hand or when desperate for

points, and fold everything else. Even with no common safe tiles, we should try our best to not

deal in.

2.3. DEALING WITH RIICHI 119

Cutting a tile dangerous (10% danger level) against one Usually riichi with a ryanmen,

and fold with a bad wait that’s cheaper than 5200. Similarly for a tile that’s dangerous at the 5%

danger level against both opponents.

Folding against 2 riichi

We should of course prefer common safe tiles, but instead of cutting tiles that appear safe against

both (no suji terminals), we should cut tiles that are conrmed safe against one opponent.

We should emphasize not dealing into the opponent we don’t want to deal into (dealer or someone

who threatens our placement). While defending against 2 riichi is stressful, nding the safe tiles

against the more threatening opponent, and then choosing the tile that looks safest against the

other opponent makes it easy to defend.

What to do in borderline situations

We’ve listed some situations as clearly good to attack or defend, others as borderline. What factor

should we look to in these borderline situations?

Point standing and placement situation

How many safe tiles we have (folding success chance)

With 3 tiles in the mid game or 2 tiles, we can usually fold successfully. But what should we do if

we can’t?

• 0 safe tiles Can’t help but attack.

• 1 safe tile in the mid game We should also usually attack, unless we especially want to

avoid dealing in or would have to push an especially dangerous tile.

• 2 safe tiles in the mid game, 1 safe tile in the late game Attack in borderline situations

and fold otherwise.

Since it’s more dicult to defend with an open hand, we should be a bit readier to push with

open hands in sketchy situations.

Other players’ strategy

If other opponents are pushing dangerous tiles, we should fold more often. A ron between them is

more likely, and many safe tiles will be created allowing us to defend more easily.

When in iishanten

1-turn tenpai chance

A rough formula for the chance in in percent of reaching tenpai in 1 turn is

p = tile acceptance ×

120 CHAPTER 2. PUSH-FOLD JUDGMENT

Where 5/6 is an approximation of 100 divided by the amount of possible draws. While this

will depend in reality on discards and the tiles in opponents’ hands, it’s a good rule of thumb for

thinking quickly.

Tenpai chance 10%, guaranteed good shape

A hand like weooofgnnm,.j. (Strictly speaking, the tenpai chance is higher

than 10% if we can see less than 4 tiles of qrdh.) According to Scientic Mahjong, if we push

everything with this kind of hand, deal-in rate will exceed win rate, by up to double in the late

game. Even with an 18000 point hand, preserving iishanten isn’t worth it if we would have to push

a tile at danger level 7%. We should understand that tenpai and iishanten dier like heaven

and earth, and a double ryanmen iishanten is not particularly wide.

However, we should in fact attack with a hand like this in the mid game if we have non-dealer

conrmed haneman (12000) or better. The reason is that conrmed haneman is often either a big

dick chinitsu or a hand with many dora. Since chinitsu is easy to call which increases tenpai chance,

and having many dora means opponents will have less, it’s ne to attack.

20%

A hand like rtyuissdm,..., or a double ryanmen iishanten where we can call.

Win rate and deal-in rate are roughly similar, so we can push a danger level 7% tile with a mangan.

>20%

Common with headless or sticky shapes. Attacking with these while still not in tenpai can be quite

good, and we should treat them roughly as we do tenpai with a bad wait. This means all-out attack

with dora dora, and folding with no added value.

Callable hands

Since a tenpai is much better than an iishanten to counterattack with, we should call everything

possible if an opponent has declared riichi. (Unless we really need a big dick hand or we intend to

fold anyway if we don’t get a good draw.)

Example ertdffggkkvb4

While this hand is easily haneman closed, and we would try to complete it closed in an equal

scenario, if someone has already declared riichi, we should call into 1000 open tanyao.

<10%

Cases where it’s good to attack from a narrow iishanten (especially chiitoitsu iishanten) or 2-shanten

are extremely rare. The main exceptions are wide, expensive and easy to call 2-shanten (chinitsu

etc.) where dealing in would not change our placement.

2.4. DEALING WITH OPEN HANDS 121

2.4 Dealing with open hands

Characteristics of open hands

We’ll now discuss defending against open hands. Even against these, defending according to the

danger level diagram is eective. However, there are some unique points of judgment.

We don’t know if they’re in tenpai Except with 4 calls (hadaka tanki), we can’t be sure

unlike with riichi (in rulesets that disallow no-ten riichi :^). 3 calls in the early game (6th turn or

before), 2 calls in the mid game (around the 10th turn) and 1 call in the late game (13th turn or

later) are roughly 50% chance of tenpai. 3 calls in the mid game and late game are respectively

about 80% and 90%, 2 calls in the late game are about 70%.

Since cutting a tile of danger level 10% against an opponent who is 50% likely to be in tenpai

is the same as cutting a tile at danger level 5% against riichi, we should weight the relative values

of our and the opponent’s hand and decide accordingly. However, in practice estimating whether

an opponent is in tenpai will depend greatly on his discards, so we should exercise caution when

using general criteria like these. We’ll return to this subject in the chapter on reading.

The score is easier to read It’s hard to read how much a riichi is worth, and its value can get

inated by ura dora, so in the previous section we used a at estimate of 7000 (dealer 9800) points.

However, we can use reading skills to gauge an open hand’s value and decide to push or fold with

greater precision.

The wait is easier to read Often when a yaku like honitsu or toitoi is obvious, we can safely

exclude many tiles. Conversely, this makes tiles that can be targeted more dangerous than when

defending against a riichi, which can be easy to overlook. (There are many players who know how

to fold against riichi but have low awareness of open hands.)

The tiles after the opponent’s last tedashi are safe Even if an opponent has done tedashi,

tiles from several turns ago become quite safe against shanpon, and tiles cut by his kamicha from

several turns ago become quite safe against sequence waits too. It rarely happens that he was

unable to call to advance his hand and then drew the wait later. While it’s not needed to remember

all the tedashi and tsumogiri, it’s good to remember them for an opponent who has called. (Also

note that an opponent might decline ron to target us in particular, though this is very rare.)

The outside of early discards (especially neighbors) becomes safer than against riichi

Cutting w from wwe early is especially rare with an open hand.

The neighborhood of the tile that came out after the last call (sobaten) is dangerous

A perfect iishanten is common with open hands as it’s easiest to call into, and the tile that comes

out to make tenpai will often be from a paired joint to conrm the joint. However, calling from a

See https://blog.kobalab.net/entry/20180203/1517667551 for a table based on actual data

Tedashi refers to cutting tiles from the hand and keeping the draw, as opposed to tsumogiri, which is to

immediately discard the drawn tile.

122 CHAPTER 2. PUSH-FOLD JUDGMENT

paired joint and cutting the overow tile (eerchii wcut e) means another area is dangerous,

especially where no tiles have been discarded.

The area where no tiles have been discarded becomes more dangerous

Since shanpon waits are more common, the danger of live tiles

increases

Tenpai estimation

While the percentages for tenpai chance for a given turn and call count give a rough estimate, this

can change heavily depending on how many middle tiles an opponent has discarded. Since we can’r

get an exact chance, the decision to push or fold will depend to some degree on our own speed and

the opponent’s value. (But even against an opponent with 2 calls, we should fold with a narrow

2-shanten.)

If an opponent discards an especially useful tile (what this means depends on what yaku he’s

making, for example dora, suited tile in honitsu etc.), he’s likely to be in tenpai. If he cuts a very

useful tile, followed by a safe tile (dead honor), he’s especially likely to be in tenpai. In this case, a

ryanmen is likely, and the outside of the useful tile becomes a lot safer. (We can infer an extra tile

iishanten earlier.) This technique is of more limited use for yaku other than tanyao and yakuhai.

If other opponents are cutting dangerous tiles against a threatening open hand, they’re also

more likely to be in tenpai. If their hand is open too, they’re probably also in tenpai. Furthermore,

an opponent who declares kan is usually tenpai or 1-shanten.

Defending against specic yaku

Tanyao

Will be very common. Usually easy to read with an opponent cutting terminals and honors early

and holding on to middle tiles. When playing with aka, we should always be aware of how many

are still unaccounted for, since the opponent might be holding all of them. Dealing honors and

terminals is an easy defense against open tanyao, but sometimes we’ll get tricked by what was

actually yakuhai atozuke or concealed yakuhai. The following are often signs of a tanyao actually

being yakuhai:

Discarded terminals after middle tiles Might have been holding them as a safe tiles, but gets

less likely the more calls he has made and the more terminals discarded late.

Called a 23 or 78 ryanmen after having discarded the 5 earlier In real tanyao, it’s better

to discard 2 or 8 to avoid furiten.

To deal with this, if we’re going for yakuhai atozuke that could pass for tanyao and draw n

with m,. in hand, we should cut n. Rather than hope to draw /, it’s better to try to trick

opponents.

Westerners often misuse this to mean “dangerous tiles”, but it means tiles that have not been discarded by

anyone yet

This obviously depends heavily on meta and general skill level :^)

2.5. SUPPLEMENT 123

Yakuhai atozuke If an opponent has made any calls with terminals, this is the most likely reason

especially if other yaku are improbable. We should be aware of the number of live yakuhai left. It

can sometimes be good to choke live yakuhai (or fold if tenpai is likely), depending on how much

we care about the other 2 opponents advancing.

Sanshoku and ittsuu When 2 groups have been called, a wait for the 3rd one is likely.

Honitsu Should be fairly obvious to notice (o suit middle tiles and ryanmen in discards early,

yakuhai late). If he has discarded a suited tile or a live yakuhai, 1-shanten or tenpai is likely. If

not, tenpai is unlikely and we can often get of rid of suited tiles before it gets too late.

Toitoi Middle tiles of all 3 suits discarded early, possibly ryanmen dropping, yakuhai only late.

Live tiles become especially dangerous, while dead tile are safe.

Chanta 456 discarded early but no 3 or 7. 456 will pass.

Other yaku should not be that hard to defend against as they tend to be quite obvious. If we

can’t win realistically, we should fold from the mid game on against a decent open hand, cutting

tiles that invalidate yaku if have no safe tiles.

2.5 Supplement

Push-fold judgment just before a draw

Just before a draw, the chance of winning the hand is low, so score and a good wait become less

important and the danger level of the tile we would have to discard becomes more important.

We can estimate the danger level by counting the number of remaining suji and by other reading

techniques.

Assume there’s one opponent in tenpai, and pushing a tile would certainly let us receive no-ten

payments, while dealing in would cost a non-dealer mangan. Then we should push when the chance

of dealing in x satises

1500(1 − x) − 8000x > −1000

that is when

x <

= 0.263...

And similarly when x <

= 0.185... when pushing against a dealer mangan. This is unex-

pectedly high, but it assumes that we are certain of receiving no-ten payments, so if multiple turns

remain we should push only at a lower danger level. Furthermore, since the number of remaining

suji just before a draw will be low, we must bear in mind that even a single no suji tile against 2

tenpai opponents will be very dangerous.

This decision is heavily inuenced by point standing. Close to all last, it’s good to emphasize

expected placement over expected round value. For example, if we are last in South 3 and no-ten

payments would raise us to 3rd place, we can risk even a no suji tile, while if dealing in would drop

us by 2 or 3 places, we should usually fold.

124 CHAPTER 2. PUSH-FOLD JUDGMENT

Rolling

In mahjong, attacking maximally when we attack and defending maximally when we

defend is often the right answer. However, it can sometimes be valuable to attempt to preserve

the hand towards tenpai or a win while defending (especially in the late game for a formal tenpai)

if reasonably possible. We call this style of play rolling (mawashiuchi), or choking when dealing

with no-ten opponents.

When attacking maximally, tile eciency will tell us the best answer X, but sometimes cutting X

won’t be reasonable (yet) from a defensive standpoint. In that case, deciding which of the possible

less ecient moves to make is the skill of rolling.

When to roll

First, we should use the basic defensive criteria to decide whether it’s worth it to attack maximally.

If it isn’t, we can cut a safe tile that still advances the hand is some sub-optimal way if completing

the hand is possible. Often, when we have multiple isolated unsafe tiles against a riichi, it won’t be

possible to complete the hand anymore and we’ll have to fold. In practice, opportunities for rolling

will be few, and we’ll fold in the majority of cases.

How to roll

If we have to break up a group, the hand will almost always be fucked and we’ll have to fold, so we

should avoid that if at all possible. On the other hand, if we have a safe oating tile, we can cut

it without a problem: we won’t usually call that rolling, instead referring to the following kind of

plays:

Cut a tile from a complex joint Since this hurts speed the least, this is the most commonly

used. We use it when we have a tile that we don’t want to risk right now, but would be okay risking

once in tenpai. (If it would be too dangerous even in tenpai, we can’t use this method.) If we would

draw a pair or joint into the unsafe tile, we would drop a dierent component; if we would draw

another unsafe tile, we’d fold.

Dropping a pair or joint When dropping joints, we of course prefer to drop those of which

both tiles are safe. While this is more tile eciency, when deciding which joint or pair to drop in

a non-rolling situation, we should often keep the ones that opponents would be likely to call.

Dropping a set Slightly less bad than breaking a sequence since we get left with a usable pair.

Often amounts to folding.

Aiming for chiitoitsu Even with multiple unsafe tiles, we can always draw a lucky chiitoi. We

should therefore preserve this possibility, but often this will amount to folding.

Rolling is an extremely high skill cap low return skill that relies greatly on mastery of all other

mahjong skills. For a demonstration of rolling in practice, watch some pro games on Abema.

2.5. SUPPLEMENT 125

Ippatsu disruption and haitei shifting

We can also make unexpected calls with another purpose than winning and formal tenpai. We can

disrupt the ippatsu of an opponent who has called riichi or shift the haitei to a more favorable

player.

If we can call to disrupt ippatsu while advancing the hand, we usually should. Of course, if we

have a chance of winning which a call would negate, we shouldn’t call. If winning is hopeless, we

should call if we are condent in being able to successfully fold after (we have enough safe tiles).

In the late game, we should start being conscious of who will draw the haitei. It goes without

saying that we should only consider calling to shift it if we have enough safe tiles.

However, there are also cases where we should intentionally not make these calls even when

we could. For example, if we’re in a close placement duel with the dealer, so that an opponent

uninvolved in this duel would improve our placement by tsumoing a big dick hand.

Chapter 3

Reading

Principles of reading

Types of reading

In mahjong, there are several kinds of “reading”, but the most important ones (the ones with the

biggest inuence on results) are

• Wall reading (how many tiles of a type are left in the wall)

• Yaku reading (guessing which yaku an opponent is aiming for, as well as hand progress reading

and score reading)

• Wait reading (how likely a certain tile is to be targeted by an opponent)

All of these depend on guessing what opponents hold in their hand from their discards (what

they don’t hold must be left in the wall). They can inuence both tile eciency (what tiles are

easy to draw) and push-fold judgment (how dangerous is this tile, how expensive does their hand

look).

Things to keep in mind

The foundation of reading is estimating the possible combinations of tiles And only

then the situation and the intentions of the opponents. For example, a sequence is much easier than

a concealed set, so at any given point, an opponent is much likelier to hold qwe than 555.

Furthermore, since there are always more unseen tiles outside of a particular opponent’s hand than

the 13 in it, it’s clearly less probable that a certain opponent has a pair of 5 than that none of

them have it.

Of course, if an opponent is going for honitsu or toitoi, or when 5 is dora, they become more

likely to hold it since they will keep it longer, so the probability of someone holding a pair or set of

5 will increase. However, since it’s statistically more likely to draw a joint or sequence, those will

be still be more likely.

127

128 CHAPTER 3. READING

It’s dicult to read rare cases (Instead of reading them, we’ll have to read what’s not the

case and deduce form there.) For example, it’s very dicult to specically pinpoint a riichi’s wait,

but reading what’s not the wait is trivial (genbutsu always pass). The probability of a particular

tile being the wait is much lower (usually 1/34 or 2/34) than it not being the wait (33/34 or 32/34).

There are some exceptions to this. For example, a player who needs to score 3500 in all last and

is in hadaka tanki with only a haku visible for score is obviously waiting on dora :^) Similarly, a

player who’s pushing all no suji tiles into a riichi is obviously in damaten for a big dick hand. But

being able to read rare cases with a high certainty is also rare.

We shouldn’t try to read where reading wouldn’t help Due to the nature of mahjong (high

amount of concealed information), situations where reading is eective to a high degree are rare.

Furthermore, often even being able to read won’t change our decision. For example, if we can read

a riichi’s hand accurately, but our hand is fucked, we’ll fold anyway. Often, reading incorrectly is

worse than not reading at all. In general, we should evaluate the situation from the certain

information, and fall back to uncertain information only when we aren’t sure.

3.1 Counting

The 1/18 rule

In mahjong, we’ll rarely be able to pinpoint waits exactly like Akagi Shigeru. In the majority of

cases, we’ll be able to exclude some possibilities, and work from there to get a probabilistic estimate

of the remaining possible options. The single most eective reading technique is counting the

number of remaining suji. When guessing an opponent’s wait, we’ll use this as the base. Since

there are 6 suji (14, 25, 36, 47, 58, 69) in each suit, there are 18 in total.

If there are 12 suji left against a good wait riichi, a no suji s will be targeted 1/12 of the time,

but a no suji b will be targeted 1/6 of the time (since 456 are each on 2 suji).

Of course, this is only a rough guideline. In practice, there will be other waits than ryanmen,

so we should treat these numbers as a high estimate. If pushing would be good even with a high

estimate, it’s denitely good to push. Conversely, since about 65% of riichi are ryanmen, the danger

level of no suji t can be estimated at 13/120 = 10.83% for ryanmen alone (low estimate).

If there are especially likely non-ryanmen waits (dora yakuhai, live yakuhai against honitsu,

especially suspicious suji traps), we can add them to the count. For example, if an opponent is

tenpai for manzu honitsu, and there are 2 safe suji and 3 unsafe honors, the danger level of no suji

w is 1/7 and of no suji t 2/7. We can understand from this example that when the amount of

possible waits is low, dangerous tiles become especially likely to deal in, which we should remember

when defending.

Honor waits

The following table gives a danger level of honors based on data of riichi between the 9th and 14th

turns:

Danger level of honors

3.2. DISCARD READING 129

yakuhai

cut held danger

0 1 7%

1 0 4%

0 2 4%

1 1 2%

2 <1%

cut held danger

0 1 5%

1 0 4%

0 2 3%

1 1 3%

2 <1%

guest wind

What can we learn from these data?

Against a late (9th turn) riichi, usually at least 5 or 6 suji will be safe. If there are 12 suji left,

the danger level of of 456 can be estimated as 1/12 = 8.3%. This is higher than the deal-in rate of

7% for live yakuhai. Therefore, live yakuhai are safer than no suji number tiles.

However, 7% is quite a lot, so we shouldn’t cut them from a hand that has a low chance of

winning. Even if it would destroy the hand, we should fold with another honor or a suji terminal.

Even the more because dealing in with a yakuhai tends to be more expensive than with a number

tile.

Honors that the opponents can see discarded are safer than those we hold in hand.

This is because calling pon on the rst one that comes out is common, and players tend prefer waits

they can see less of.

Furthermore, yakuhai are safer than guest winds at 1 held/1 discarded. This is because

of the aforementioned yakuhai pon, and because guest winds are classic choices for tanki.

3.2 Discard reading

Types of discard information

We’ll now start with the reading of particular discard patterns. The following kind of specic hints

are most often useful:

• Calls Since these are part of the hand. Since we can see what specic tile was used to enter

tenpai with a call, we can read more precisely.

• High value tiles cut early

• Low value tiles cut late With value, we mean ease of making sequences (middle tiles are

high value) and tiles that help give a higher score (dora etc.)

• Tedashi of joints and pairs We can understand that the opponent has more valuable joints.

The better the dropped joint, the more information we get this way. For example a riichi that

dropped a penchan can be almost any wait, but one that dropped a ryanmen is likely to wait

on a ryanmen or better. While dropping groups is rare, we can understand that the opponent

is going for an unusual hand.

• Other rare cases We should always be on the lookout for sketchy things going on. For

example, an opponent who pushes dangerous tiles against riichi is probably in damaten,

especially if also in the lead.

130 CHAPTER 3. READING

Sobaten riichi

A long time ago, the theory that “the tiles around the riichi tile (last tedashi in the case of a

tsumogiri riichi) are especially dangerous” was widespread. This theory was disproved in Scientic

Mahjong.

The problem with this theory that the single riichi tile is not nearly enough information to be

able to deduce a particular pattern. We can think of many possible hands where the wait will be

elsewhere than around the riichi tile. To recapitulate from the section on iishanten theory, there

are four main types of iishanten:

1. Example qqqyussfghjcv

2. Example qqqyussfghccv

3. Example qqqyusdfghcvb

4. Example qqqyssfghjcvb

In all of the above, we are tenpai if we draw t. However, we’ll only discard a tile around the

eventual wait in example 2. Furthermore, even in this example, we won’t sobaten riichi if we draw

sxcbinstead of ti. (The probability of sobaten would be even lower of yu were yi.)

I hope we have made clear how low the success rate is of the sobaten riichi theory. In the majority

of cases, we can’t do any better than to count the number of remaining suji.

However, if we assume an iishanten of the appropriate type, we do get some information about

the wait. If we suppose that the probability of such an iishanten is 100% when an opponent has

called riichi with c, then the area around c will be targeted about 50% of the time. (The other

50% the wait in the area will have entered rst and the target will be the other wait.) Taking into

account all the possible shapes with a c like zzc, xcc, ccv, cbm etc. we can

arrive at a danger level for z of about 20%. This is considerably than the roughly 8.3% we’d

expect from counting suji, so the theory can be said to be eective. If the opponent has called and

then discarded c, the probability of sobaten will be double, since we know the other area entered

rst. So how often will an opponent have this kind of iishanten? We’ll talk about that next.

Gyakugiri

One of the cases where a perfect iishanten is likely is when a high value tile has been cut early

and a low value one late.

With an extra tile iishanten, it’s normal to cut a high value tile that’s useful for upgrading the

wait late. (the exception are hands with 2 ryanmen that want to keep a safe tile.) The same is true

for sticky iishanten. If the probability of these shapes is low, a perfect shape is more likely, with

for example a late q discard likely to be from qqw qqe qet etc.

3.2. DISCARD READING 131

For example, let us consider the opponent who called riichi after discarding g→s. If it wasn’t

a perfect shape, what could it have been? In an extra tile or sticky shape, it makes no sense to

keep s over g unless going for sanshoku or chanta, or holding sdd hoping to upgrade into

a wider headless iishanten with af. We can hardly imagine an extra tile iishanten where both

sg were useless, since he would have discarded s very early. In a headless shape, we can

imagine sssg, cutting g→s to make s the head, or asg/sdg cutting g rst

and drawing a/d to make the head (rare). It is also possible that he cut g before another

opponent called riichi, stopped on s for a bit, then rolled into tenpai and risked s.

What should be observed here is that all of the above are rare cases, and the simple case of a

perfect iishanten will be more likely unless we have extra clues.

We can easily imagine this discard order from aasg or ssdg (very rarely from

sddg when aiming for tsumo sanankou), so that f and especially paia will be more danger-

ous.

The deal-in rate of a will be

(1 − rare cases) × (1 − irime)

Since the probability of irime is about 50%, the deal-in rate of a will be about 40% assuming

20% chance of one of the above rare cases. Since ssdg will be more common than aasg

(it’s common to cut srst from aasgbefore iishanten), the deal-in rate of fwill be roughly

25%. While this is a back-of-the-envelope calculation, pushing a tile at that danger level only makes

sense with a good wait mangan tenpai.

If an opponent discarded g and then called elsewhere and discarded s, we can eliminate irime

and the danger level of a will be even higher.

This kind of g→s anti-ecient discard order is called gyakugiri. With gyakugiri, the area

around the second tile becomes much more dangerous. With other patterns, we’ll be able to adapt

the above idea, but there will be more exceptions, so we should remember that 52 and 58 gyakugiri

reading is the most eective.

Patterns of gyakugiri

g→s with other tedashi before declaring riichi The probability that the area around s

has been completed becomes higher. If we think that the probability of the wait being in that area

when s was cut at tenpai is 1/2, we can think that it’s 1/3 (3 incomplete joints left to complete)

when s was cut at iishanten. However, we can’t exclude other iishanten types. As the number of

tedashi afterwards increases, the probability decreases further. Therefore, we can’t really consider

this kind of reading more eective than counting suji.

132 CHAPTER 3. READING

t (middle tile from other suit)→s We can add the cases of tsdfgsticky iishanten,

ittsuu, honitsu which won’t make a big dierence, but instead of just aas and ssd we

can imagine shapes like ssf, sff, sfh and rarely sfghhj. The danger level

of a will thus be about 30–50% that of the g→s case (15% in total, about F rank), which is

still more dangerous than the 8.3% we would expect from counting suji. f will also be a bit more

dangerous, but not especially more than the 16.6% we’d estimate from counting.

h→s Instead of just aas and ssd we can imagine shapes like assh (suji trap)

and ssfh. The rest is similar to g→s. awill have a danger level of about 20%, f won’t

be especially dangerous.

hrightarrowd When playing with aka, cases with sticky iishanten will be unlikely, but we can’t

be as sure as with g→s. Also, it’s not just af but also sg that become more dangerous.

Since fg will generally be ryanmen, and s will also usually be targeted with ryanmen (cutting

a from add for a suji trap), it’s only a that becomes especially more dangerous.

(middle tile from other suit)→d We can think of numerous exceptions, so reading is not

very eective. a will be slightly more dangerous.

Summary

• With g→s riichi, a is especially dangerous, f is more dangerous than F rank.

• With h→s riichi, a is about F rank or a bit worse.

• With middle tile of another suit →s riichi, a is more dangerous than normal but not F

rank.

• With d riichi, a is more dangerous than normal.

Call reading

We’ve discussed before how a perfect iishanten with paired joints is more ecient with an open

hand. However, we can’t exclude other types. Here, we’ll discuss some examples where we can infer

that an opponent has a perfect shape.

3.2. DISCARD READING 133

This is a tonpuusen sudden death overtime with 30000 to win. Blue lines indicate tsumogiri,

the tiny illegible insets are (from top to bottom) 7m chii, haku pon, hatsu pon. Source is a dead

mahjong BBS post from 2008.

While we want to cut b into tenpai, toimen’s call is worrying. What should we do?

Calling chii on u and cutting v certainly makes the area around v dangerous. However, we

can’t be sure he had a perfect iishanten. Since winning this hand will put us in 1st, we should risk

it and push b. However, this hand is indeed a perfect iishanten, since he cut t from tpy,

and it would make no sense to keep a v if it was being used as just a oating tile. So he must have

had something like xxv, cvv and vvb (accounting for the fact that he cut m and

2 c are visible). Here, we can eliminate vvb since he would have called chii on the dealer’s

discards. So the remaining possibilities are xb ryanmen and x + something shanpon. Since

cutting b has an extremely high chance of dealing in and dropping us to 3rd place, we should cut

n preserving tenpai defensively.

Summary: opponents who cut from a paired joint are likely to have a perfect shape

(other paired joints)

10hoe looks better nowadays :^)

134 CHAPTER 3. READING

Dora discarding

Dora discarding riichi

Unless the hand is suciently expensive (chiitoi with 2 more dora, menhon chiitoi etc., all very

rare) a tanki wait is very unlikely. Furthermore, the chance of a ryankan suji trap becomes

lower unless the hand is already expensive enough. However, we don’t really know anything else.

The reason is that a dora is so valuable, almost all closed hands naturally want to keep it until very

late if possible, even isolated. Since the information value of keeping a dora until late is low, we

can’t use it to deduce a certain shape.

While “suji that cross the dora are safe against dora discarding riichi” is a proverb, this is not

actually reliable enough to make these suji safer than other untried suji.

Dora → something else riichi

Conversely, in this case we have more information to work with. It’s easy to understand that

a dora-related wait becomes less likely. Exceptions are hands like werdffxxvnn

dora f or weeyui;hhxcv,, dora h keeping the dora worsens the shape and

makes it harder to make yaku. We can’t really read these, but they’re rare cases. This technique

becomes more powerful with a more central dora. For example, i dora discarded makes yo

ryanmen, uo shanpon, u penchan as well as tu kanchan safe, while g dora makes all pinzu

waits except penchan, askl shanpon and afhl ryanmen fairly safe.

Furthermore, we can read that a perfect iishanten is common from the following reasoning:

• Dora is the strongest oating tile, so an extra tile iishanten is unlikely.

• Same for sticky iishanten. While an edge dora is hard to stick to, discarding a middle tile

dora is especially rare.

• Headless iishanten is also slightly less likely. For example, in qqwedora q, we’d rather

conrm a dora pair than cut q into headless iishanten.

However, other types can’t be eliminated. Furthermore, if combined with gyakugiri to make

dora g→s, we can be almost 100% sure of a wait around s. The only notable exception is

headless sssg.

Example 3k5.p;

c dora u

If the area around i isn’t irime, we can read that the wait is either yo or o + something

shanpon. So the danger level of o is about 50% and y about 25%. We should only consider

cutting o with a good wait mangan tenpai. (Similarly for dora h→d.)

(If you’re curious, the hand in the actual game was qweuioozxcjkl.)

3.2. DISCARD READING 135

Joint dropping

What we can read from joint dropping

An opponent who drops a joint is holding on to a better (rarely equally good) joint

unless he’s folding

If an opponent enters tenpai immediately after dropping a joint, he doesn’t have

a headless or sticky iishanten (The exception is dropping ew from wwwe or a sticky

shape going from tenpai into a better tenpai.)

Before reading any more, we should remember the above points. If an opponent drops a joint

before iishanten, he might also be holding a oating tile instead of a better joint. Of course, the

better the dropped joint, the more we can infer.

Types of joint dropping

Penchan The weakest joint, so gives the least information. The hand is more likely to be close

to tenpai when dropping from the inside out (w→q) than from the outside in.

Kanchan Doesn’t have a penchan unless it gives dora or yaku. When cutting e→q, he doesn’t

need an upgrade to ryanmen with r, so either has a qety shape or all ryanmen left. We

can usually expect a ryanmen end shape (or a headless shape tanki). When cutting from the inside

out, a wt wait is hard to imagine so w will usually pass (and t too if i is safe). While there

are exceptions related to dora and yaku, there will be less traps than with suji, so the tile that

completes an outer kanchan dropped from the inside out is safe.

Ryanmen We can expect a ryanmen end wait (or headless shape tanki). If we know for sure an

opponent has a ryanmen, suji and honors are safe. Therefore, the danger level of no suji tiles

becomes equal to 1/(number of remaining suji).

While this isn’t pair dropping, an opponent who declared riichi with a safe tile will also usually

have a ryanmen (especially in the early game), since with a bad shape it’s often good to keep a

oating tile for upgrading the wait.

Exceptions

While we say that an opponent who drops a kanchan from the inside out usually has a ryanmen or

a dora/yaku-related bad wait, there are exceptions. With r→w, a shape like qqwr into q

+ something shanpon is possible. Especially with r→ something else → w, a simple kanchan is

hard to imagine. q is suji, but if a rw kanchan has been dropped this way it’s as dangerous as

no suji, since it’s hard to draw into qqw and irime will be rare. Conversely, with w→r, q

will usually be safe.

When choosing which ryanmen to drop, there are some that are preferable, for example overlap-

ping shapes like ssdd and sdgh. If an opponent has dropped sd, we can be liable

136 CHAPTER 3. READING

to think that af are safe, but this is not true. Especially with some other discards in between

s and d, af become very dangerous.

To summarize, if an opponent drops a joint with some time in between its two tiles, he’s usually

holding on to another shape in the area of the second tile (else he would prefer another oating

that doesn’t create furiten).

Tedashi ryanmen → something else riichi

We can infer that the opponent dropped the ryanmen with too many joints before iishanten. There-

fore, the tile that came out after must probably have been connected to another ryanmen somehow.

Suppose that it was u, then yuu and uui are the most likely candidates, making ti

and yo dangerous. While a shape like rrttyu into ey wait is also possible, this is of

course much less likely. With a dropped kanchan, sobaten also becomes a bit more likely in this way,

but there will be many more exceptions (oating tile for upgrades etc.). With penchan dropping,

it’s normal to keep a normal middle tile, so we can’t really deduce anything. If the dropped joint

contained or accepted dora, we can be especially sure of the read.

Dropping a dora bad shape joint

We can infer than the opponent has a better joint, so either a yaku-creating bad shape or most likely

a ryanmen. When cutting /→c, a shape like xcc/is common, so zvare dangerous. The

other possibility is a big dick hand that cuts / rst for safety. When cutting c→/, xcc/

into zv wait is also suspect.

Dropping a dora ryanmen

When dropping a dora-accepting ryanmen, the most common reason is an overlapping ryanmen,

with a dora-containing ryanmen a double overlapping ryanmen. The exception to both are big dick

single suit hands. When cutting f→; in 3 aka rules, a shape like ffg; can be eliminated

(though there are some players who will do this with a conrmed mangan to trick opponents).

Outside of an irregular hand, we can think of a headless iishanten that entered ; dama tanki and

then declared riichi after changing the wait. If the dora was discarded rst, we should be especially

wary as a big dick irregular hand is likely.

Pair dropping

What we can read from pair dropping

A tanki wait is unlikely

A headless or sticky shape is unlikely

The opponent is holding a better (or rarely equally good) pair

3.2. DISCARD READING 137

Pair dropping with an interval between the two tiles

With an honor pair, we can think of keeping the second tile as a safe tile (so ryanmen extra

tile iishanten is likely), but with a number pair, the second tile most have had some purpose.

While ukase-uchi is possible, situations where it’s ecient are rare, so we can assume the tile

was connected to another pair or joint. For example when dropping e→ something else → e,

shapes like qqee, eett, eetu, eetyuui are possible, while shapes

like eeer or weee will be uncommon, so qr and wt will be hard to target and

especially w will be safe.

Honor pair A ryanmen wait is likely. The reason is that with a bad shape joint, it’s common

to prefer the honor pair aiming for honor shanpon. When dropping a yakuhai pair, pinfu is likely.

Dropping a double yakuhai (renfonpai) pair is a surere sign of a big dick pinfu (menpintan dora

or worse) so we should be very careful.

Aka containing pair Either a big dick hand with yaku or likely to have an omote dora pair

with ryanmen wait or better. It’s common to prefer an aka pair over a bad shape joint, so a

ryanmen is likely. We should also be careful.

Dora pair Either a honitsu, chinitsu, yakuman or chiitoi with the other 2 dora. Needless to stay

we should be extremely careful.

This kind of yaku reading becomes more eective in open hands. If an open who’s going for

toitoi drops a pair, he has a better (easier to call or yakuhai) pair. If a calling opponent drops

a yakuhai pair, he’s usually going from yakuhai atozuke into open tanyao or from honitsu into

chinitsu. An opponent who drops a renfonpai pair has a big dick hand. Unless it’s the late game

and he wants to make even a cheap tenpai, tanyao dora 3 or chinitsu is likely. If a clearly expensive

call has been made and we’re far from tenpai, we should usually fold.

Yaku reading

So far, we’ve focused on reading the wait of a riichi under the assumption that opponents will play

eciently to maximize speed. In this case, aiming for yaku is more of an exception. We’ve already

given a short overview of how to tell which yaku an opponent is going for in the previous chapter

as it’s not that dicult.

We can use discard reading on opponents that appear to be going for even very rare yaku. For

example, suppose an opponent has called pon on 1 and 2, and then dropped a pair of 6, then

we can be quite sure he’s going for shousuushii.

However, reading the yaku of a riichi is a problem. We can read common yaku like pinfu and

tanyao being likely when the opponent drops an honor pair, but reading sanshoku or ittsuu is almost

impossible. We can at best read that an opponent who has dropped a ryanmen usually won’t be

going for a sanshoku or ittsuu bad wait.

If an opponent has many middle tiles in his discards, the probability of an irregular hand does

increase, but we can’t eliminate that his haipai could have been very good (which is quite common).

To read an irregular hand, we need evidence that the opponent did not advance straightforwardly.

138 CHAPTER 3. READING

Example South 1, dora s (source: Sasaki-pro’s hand in an online tourney)

.hnm

(all tedashi)

The point here is that he dropped nm to keep 3. If 3 were a number tile, we’d expect

sobaten around it, but we can only imagine a regular hand advance with cutting 3 from an anko

headless shape. Since we can assume it was used to make yaku, we can narrow it down to manzu

honitsu and chiitoitsu. In this particular game, the dealer, Fujisaki-pro, risked a u with a chiitoi

dora tanki and dealt into a haneman (riichi honitsu double south). While with a dealer chiitoi dora

dora, it might be worth it to risk it, with a cheap hand it should be no problem to fold even from

good wait tenpai.

We’ll now discuss some examples of how “sketchy calls” can be used to read yaku.

Example dora 7, 4 is a guest wind (source: Dahime Obakamiiko, Umasugi’s hand)

sdfghjkl445577

Here, Umasugi cut 5, called pon on 4 and cut 5. This is quite a sketchy call from the

opponents’ point of view. Since 7 is dora, they can read that she has at least two 7 and

discarded 5 because 4is easier to call and she doesn’t need any more points. While an exception

would be going for shousuushii, this is of course extremely rare. Since a 7 pair requires another 2

han for mangan, we can narrow the possible yaku down to honitsu, toitoi and chanta sanshoku.

The idea behind this play was that if afj enter rst, 4 will come out easily with the 5

pair in the discards. However, cutting 4 and calling pon on 5 can also be a haku only, or no-ten

(cutting a guest wind pair from a weak hand is common). On the other hand, the sketchy 4 pon

after dropping the pair of 5 is likely tenpai. If there isn’t a dierence in score, plays like this

that aim to trick to opponent are good (afj for the trap are more likely to come rst than 4

that’s easy to read), but here I think that the straightforward 4 dropping is better since there’s a

chance at haneman or baiman.

Tanyao with 2 calls that called

xc chii and cut m

The mainline is kuinobashi from xcvbm into a cn wait. Going for tanyao with a xc

uncertain shape, it’s rare to make 2 calls to other parts of the hand. However, even if this play

makes the wait obvious, it’s still eective to extend the wait hoping for tsumo.

l pon, then

xc chii and cut v next turn

We can read chanta, but with another yaku (otherwise, wouldn’t call uncertain lpon), usually

yakuhai or sanshoku doukou. (very rare in actual play)

3.2. DISCARD READING 139

Sobaten tanki

Sobaten can also occur in a headless shape. For example, drawing or calling chii on ru into

tygh will create a sobaten tanki. While this is very dicult to read in closed hands, and

not calling riichi while waiting for the wait to improve will be common, it’s especially easy to read

in hadaka tanki (almost certainly sobaten tanki) or other open hands with many calls. Of course,

when a tedashi is made after the call, the theory no longer works. Furthermore, we can often discern

a tanki wait by the opponent doing multiple tedashi after calling into what appears to be tenpai.

Chiitoitsu reading

Apparently, Totsugeki Tōhoku (whose real name turns out to be 作田誠, Sakuta Makoto) wrote an

article about imperfect information games and presented it at a symposium. Notably, it included a

simple program that was able to detect honor waits and chiitoitsu with higher accuracy than any

of a sample of 30 human mahjong players.

The algorithm for detecting honor waits was based in several criteria, such as

• A certain number of middle tiles discarded before 3 honors and terminals

• At least 40% of the discards are middle tiles

• At least one live honor exists

• A certain number of live or once discarded honors exists etc.

The algorithm for detecting chiitoi was based on a single criterion.

• The riichi tile is a live or once discarded honor

With simple criteria like these, a computer was able to surpass human players’ “reading skills”.

Takeaways

The foundation of detecting honor waits is counting few suji left and many honors left. The

most important factor in detecting chiitoi is that the riichi tile is a good tanki tile. We

can combine these techniques with those discussed before for reading irregular hands, for example

ryanmen dropping in discards like yu

Eective wait reading situations

We’ve discussed the main cases where reading a specic shape is more signicantly more eective

than counting alone. However, making a specic read successfully often won’t change our strategy

(for example, if we’re already folding).

Suppose there are 11 unsafe suji left, and we have read one of them to have a danger level of

50%, then the remaining 10 each have a danger level of 5%. When in tenpai, we’d stop only at the

Can’t nd a full description of the algorithm sadly

140 CHAPTER 3. READING

extremely dangerous suji and push the rest, but with a weak 1- or 2-shanten it’s not worth pushing

any unsafe tile.

Conversely, if we want to fold but have no safe tiles, we should prefer to cut even slightly safer

tiles, such as suji when we are highly certain of a ryanmen riichi, or the area around the dora if it

has been discarded.

Cases where we can eectively read an irregular hand are rare. Even AI can only read honor

waits correctly about 30% of the time. However, if we assume there are at least 2 honors left, then

the danger level of each one would be 15%, which is not much higher than a no suji number tile. We

will almost never risk a no suji number tile over an honor, and we will usually simply fold anyway

(same for chiitoi, defending as against a regular hand).

Hand progress reading

An opponent can be read to have a fast hand if he discards many middle tiles early without holding

on to terminals and honors. Similarly, an opponent going for a certain yaku is usually close to

tenpai if he discards tiles related to it.

However, these are vague and not very eective techniques, since in mahjong, knowing how far

an opponent is from tenpai is much less important then knowing whether he’s in tenpai at all. It

can maybe play a small role in deciding which opponent to keep safe tiles against (keeping the

dealer’s or the placement rival’s if no specic information is available).

Another obvious but important point to be made is that an opponent who is folding is never

in tenpai, and we can ignore defending against him. We can usually detect that an oppponent is

folding if he tedashis several safe tiles in a row, or breaks up a safe tile pair or concealed set.

3.3 Wall reading

Advantages of wall reading

• We can decide more easily when unsure about which joint to drop

• We can keep easy to draw tiles when going for chiitoi

• We can know which wait is easier to tsumo

• We can estimate what tiles opponents are holding

• We can make more precise push-fold judgments (folding in borderline situations if few tiles

are left in the wall)

What we need to pin down rst is that no matter how well we read the wall, it’s still

uncertain information. If without reading the wall a certain move is clearly more ecient or gives

better value, we should play it without hesitation. Choosing a kanchan over a ryanmen because

it’s certainly live in the wall is nonsense.

We should only use wall reading in situations where we have diculty deciding based on certain

information, the “borderline” or “dicult” hands discussed before.

We should also note that wall reading does not have a major inuence on results, and it’s wise

to make decision based on reading the opponents’ hands over wall reading.

3.3. WALL READING 141

Tile group reading

To read the wall, we need to know what tiles opponents are using to build their hands. Same as

with wait reading, we need to read “combinations of tiles”.

Please remember the following points

• We can get a rough idea of how easy it is to make a sequence with a certain tile

• We usually can’t read if an opponent is holding a pair or set of a certain tile, so

we don’t try to

Tiles that are dicult to make sequences with but aren’t discarded are more likely to be in

the wall. This applies to honors as well as tiles outside of kabe (4 tiles discarded) or tiles outside

of heavily discarded tiles. For example, if 3 u, 3 i and only 1 o have been discarded, the

remaining o are likely to be left in the wall.

If an opponent is clearly going for a honitsu or toitoi, we’re liable to think they might hold a

pair of the yakuhai we have a pair of. While with obvious yakuhai atozuke this is logical, with

honitsu and toitoi it’s still much more likely that our nal 2 yakuhai tiles are left in the wall than

that the opponent has both of them.

Suit reading

We’ll now investigate the concepts of “expensive suit” (discarded little, harder to draw) and “cheap

suit” (discarded heavily, easier to draw).

Seeing which suits are expensive or cheap is easy from just looking at the discards. Since the

opponents are more likely to hold tiles of the expensive, it’s less likely to be left in the wall and vice

versa. Since opponents will make less sequences with it, the cheap suit is not just easier to draw,

but also easier to win on and to call.

Example rrtffgkkxcvm,. → cut r if manzu is cheaper and

fif pinzu is cheaper, xing the stronger (easier to draw) joint.

We wrote earlier that a ryanmen wait that opponents might hold is better than a kanchan that’s

certainly in the wall, but how about a 5-tile wait that’s certainly in the wall? Please consult this

table (source: https://blog.goo.ne.jp/21_/e/3676af9ce718a9d97f07100ba5dd0356)

Probability of drawing a tile

n opponents might hold certainly in wall

1 14% 21%

2 26% 38%

3 37% 52%

4 46% 63%

5 54% 71%

6 61% 78%

7 66% 83%

8 71% 87%

142 CHAPTER 3. READING

Bearing in mind that we will rarely be able to know tiles are left in the wall with 100% certainty,

we can see that it’s rarely good to choose a narrower wait based on wall reading.

However, with a very wide hand, dierences in tile acceptance matter less, so we can more easily

change our decision based on wall reading.

Example yudfghjcccvbbb

We’d usually cut c or b, but if manzu is expensive we can drop yu.

Reex reading

We can also read the wall from an opponent’s discards more specically. However, this method is

not very eective. While an opponent who cut q is slightly less likely to hold w, we can think

of numerous exceptions, to say nothing of the other two opponents. However, if we have no other

factors to decide with at all, we can use the following two rules:

• An opponent is less likely to hold the neighbors of a tile he discarded

• An opponent is more likely to hold the suji of a tile he discarded

Example When choosing which kanchan to break from qeuo, we should keep qe if

many qe have been discarded by opponents.

Example When choosing the end wait from dgj, we shouldn’t choose f if many aj

have been discarded. The ease of creating a suji trap after riichi also plays a role here.

With honors, from the mid game on, honors discarded once are more likely to be

left in the wall than live ones, since opponents did not call pon on them. While if we want

to stack yakuhai, we should prefer live ones (since we want a whole set), we should prefer once

discarded honors when going for chiitoi.

While the dierence is very small, if kamicha just discarded 57, 5 is more likely to be left

in the wall, since shimocha and toimen have not yet had the opportunity to match 7 now that

it’s not live anymore.

3.4 Supplement

Disguise

Since opponents will try to read our discards, it can be good to prevent showing them important

information.

For example, we should cut w from wert so that opponents won’t know ti are safe.

Since tedashi gives out more information than tsumogiri, we usually shouldn’t karagiri (tedashi an

identical tile to the one just drawn).

There are some exceptions where it’s advantageous to give opponents more information. For

example, if we cut gfrom fggand immediately draw another g, we should karagiri to create

3.4. SUPPLEMENT 143

the illusion of dropping a pair, making a dh appear unlikely. When in damaten for a riichi’s safe

tile, we should karagiri safe tiles to make opponents think we’re folding.

Even if our hand is very bad, we can make threatening discards to make opponents wary.

It’s sometimes even good to give opponents true information instead of false. For example, as

non-dealer in all last we can try to make our wait more obvious with karagiri and hope that 1st

place will deal into us on purpose to end the match.

However, techniques like these are not very important, and we should remember to have things

closer to the core of mahjong have a bigger inuence on our play.

Gyakugiri II

An appendix to the subsection on gyakugiri.

We’ve discussed the main patterns of gyakugiri, but the same can be applied to any high value

tile → low value tile discards. For example t→o makes the suji i quite dangerous, and cutting

j late with all 4 h visible makes kl suspect. If the tile discarded into tenpai wasn’t a safe

tile, and there was no irime, then it must have been sobaten.

Another point which we should bear in mind is that late terminal discards can be signs of

upgrades or slides, for example cutting . when drawing n with m,., m. or ,.. While

we can’t conclude the wait is likely to be around that area, we can be fairly certain a sequence

exists there.

(red lines are tsumogiri)

144 CHAPTER 3. READING

Here, the late l tedashi is suspicious. Drawing h into jl is mainline, but even if that’s

not the case the opponent is likely to hold at least one j. Since that means there are only two j

left, we should fold.

Lag reading

Not as important as other kinds of reading, but still worthwhile to remember.

Lag refers to the small pause that is inserted after a tile is discarded in online clients while

players consider whether to call. While many clients sometimes add random lag to discourage lag

reading, it’s nevertheless possibly to notice the dierence with some practice. Since lag reveals

information to opponents, we should usually play with calling o if we don’t want to call.

However, we should be careful since there are many non-obvious calls like kuinobashi that are still

good to make.

While we don’t know whether the lag on a certain tile is for chii, pon or kan, it’s clear that if

it’s for pon or kan we get more information. There are many possible patterns where it’s possible

to call chii, but if it’s possible to call pon, one of the other 2 opponents must have at least a pair

of the tile. We can be sure lag is for pon if

• the tile is an honor

• it was discarded by kamicha

• it’s obvious that the discarder’s shimocha can’t call it (in riichi or clearly not holding the

required tiles)

• we’re playing sanma :^)

If there’s pon lag on a tile that we also hold, we can’t draw a second one, so we should discard

it immediately if we originally wanted to stack it. Similarly, if there’s pon lag for a tile which he

have a bad shape joint waiting on, we should break it as there’s only 1 tile at most left in the wall.

If an opponent calls riichi with a tile that was previously pon-lagged, sobaten from a perfect

iishanten is likely.

Tiles outside of a pon-lagged tile becomes a bit safer, as one-chance tiles. If we can see one more

tile, they’re no chance and comparatively safe.

Tile arrangement reading

Since players will usually arrange their hand sequentially, it’s possible to read it based on this

arrangement. While there are players who don’t arrange their at all, this makes it much easier to

make tile eciency mistakes and thus isn’t recommended.

We should pay attention to not make our hand too easy to read. By applying the following

rules, we’ll be ne

• Sort by suit, with the order of the suits and honors random.

• Sort each suit in order, but sometimes in reverse lke ewq

3.4. SUPPLEMENT 145

• Don’t put tiles that we want to call or discard at the edge

For example, if we put 55677 at the right edge, call pon on 5 and discard 6, it’s

easy to read that we have a pair of 7. We should put these tiles in the center of the hand

instead and completed groups at the edge, possibly reversed. However, if we might want to

call into our completed groups, for example pon a or s into aassd, we should put

these groups in the center and out of order, like aadss

• We should arrange joints we want to call so that there are no tiles in between. For example,

with zccvb tenpai waiting on x, we want to extend by calling chii on v. However,

exposing cb this way with a tile in between makes it easy to read. Instead, we should

arrange the tiles like zcbcv

Person reading

It’s very dicult

to form an impression of an opponent without playing many matches. It’s best to

use long-term data (especially simple things like deal-in rate and riichi rate) and not rely on one’s

own experience since it’s prone to bias. In general, we also shouldn’t change our playstyle too much

in response to opponents. A lot of so-called “occult” styles are based on bias from focusing on one

particular case too much. However, if opponents have “tells” like getting visibly nervous when in

tenpai, we should exploit those. Many people think that a digital playstyle is easy to read because

it’s consistent, playing the same move in the same situation. However, playing dierent moves in

the same situation is always a loss some of the time even if opponents can’t read us. Furthermore,

playing consistently will let us think of what to discard and what to call in advance, making it

easier to play smoothly and reduce tells.

Long form shitpost abbreviated, also note that the author is talking about competent (JP) players, when playing

against chinks and /mjg/ shitters you can get a lot of information from their rank and >reputation (watch out for

novice smurfs especially :^)

Chapter 4

Point situation judgment

Principles of point situation judgment

In many mahjong rulesets, uma and oka exist.

Accordingly, round expected value is not always

equal to hanchan expected value. For example, when leading in all last, it’s common to go dama

even when we would otherwise call riichi to secure 1st place. In general, we need to be aware of our

placement at the end of the hanchan. The amount of riichi sticks and honba, as well as shuugi can

also inuence decisions. We’ll discuss how to adapt round expected value maximizing strategies to

these rules.

Rate has no impact on strategy but its ratio does The size of the rate (how many ¥ we

get per money point) is claimed to inuence strategy by many people. At low rate, its doesn’t hurt

as much to lose, so people prefer to have fun rather than to win is one line of thought. However,

if we want to tryhard win, we want to maximize hanchan expected value, which is independent of

rate.

However, the ratio of the rate (whether the uma is 10-20 or 10-30, how much the oka is,

how much the shuugi are) does have an impact on strategy. In rules that have a big reward for

taking 1st, we can make riskier plays at 1st that might lower our placement and vice versa. Other

examples are online clients that ignore points and only record placement, shizu-uma (uma that

changes depending on how many players are over/under a certain score) and tournaments where

only certain placements advance or the sum of multiple hanchans is important.

The point situation becomes more important closer to all last This should be fairly obvi-

ous. While ideally we should maximize hanchan expected value, which is a combination of expected

end score and expected placement, this chapter will focus on maximizng expected placement as it’s

easier to understand.

What the fuck are these? Uma are extra points that 4th and 3th pay to 2nd and 1st at the end, commonly

10-20 or 10-30 (in money points: each player gets base money points equal to (his score - 30000)/1000). Oka is a

bonus for rst place, usually of 20 money points, since players start at 25000 but the end score is calculated with

respect to 30000 points. The dierence goes to the winner.

147

148 CHAPTER 4. POINT SITUATION JUDGMENT

How to think about expected placement

Placement expected value formula

placement EV =

n=1

(placement points for nth place) × (probability of coming nth)

In all last, we can use this formula to determine whether it’s better to push or to pull among

other decisions.

Placement points depending on ruleset

• 25000 start, 30000 end, uma 10-20 → 35/5/-15/-25

• 25000 start, 30000 end, uma 10-30 → 45/5/-15/-35

• MFC hanchan league → 2/1/-1/2

• MFC tonpuu league → 1/0/0/-1

• Tenhou 5th dan tokujou table → 75/30/0/-105

When playing in tourneys with a prize pool or when yakitori

or busting out prizes exist, we

can adapt the formula to include them too. Since expected placement is an important strategy

concept that’s very likely to change depending on ruleset, we should endeavor to always remember

the rules and adapt to them.

Scoring techniques

To know our expected placement, we need to be able to score hands quickly in our head.

It’s not

necessary to memorize all the possible fu values. We only need to memorize 20 fu, 25 fu and 30 fu.

Then 40 fu, 50 fu and 60 fu are double those, and 70 fu is just 50 fu + 20 fu and so on.

Point dierence calculation

• Suppose we need x points to improve placement

• Then we can ron x points o any player

• Or get a x/2 direct hit

• Or with tsumo

– Get a 4x/5 tsumo if neither we or the opponent are dealer

– Get a 3x/4 tsumo if we’re the dealer

– Get a 2x/3 tsumo if the opponent is the dealer

• This is an easy to remember 1, 4/5, 3/4, 2/3, 1/2 progression.

Extra payment to opponents after not winning a single hand in the entire hanchan

Learn to score NOW if you haven’t already. Ask in the thread for good practice programs

149

• For each riichi stick in play, reduce x by 1000 beforehand.

• For each honba, reduce by 300 for ron o another player, 400 for tsumo, 600 for direct hit.

• A similar but inverse process can be used to calculate how big of a point dierence a particular

win will make

• No-ten payments induce a 4000 point dierence with 1 or 3 players tenpai and 3000 with 2

players tenpai

How to ght in all last

After all is said and done, the “daigomi”

of mahjong is risking it all in all last.

In standard 25000 to start 30000 to end rules with 10-30 uma, improving our rank from 4th to

3rd or from 3rd to 2nd is worth 20000 points, from 2nd to 1st 40000 points. 3rd to 1st is worth

60000 points. This is more than a yakuman which doesn’t change placement.

In 4th, we can’t lose placement anymore. Therefore, situations where we should content ourselves

with being last really don’t exist. It’s annoying to the other players, but in the long run, settling

for last is like strangling our own neck.

What we should pay attention to is that relying on luck to improve placement and then

being unlucky is no problem. There are many players who hate to rely on ura dora for a

comeback, or get disappointed when the ura doesn’t come.

But please consider the following. Mahjong is a game of chance. When making our hand, we

rely on the luck of the tsumo. It would be funny to treat ura dora and ippatsu dierently. Because

ura dora exists, we must take it into account. Should we rely on ura dora or take the hand that’s

certainly expensive enough? We should compare the probability of improving placement and decide

from there. If we’re missing 1 han to a certain comeback hand, there are quite some cases where

we can get it:

• At least one ura dora (roughly 30% of the time)

• Ippatsu (roughly 10%)

• Direct hit

• Kan into dora or extra fu

• Another opponent declares riichi, giving us one more stick

All in all usually at least 40%, often more than 50%. Is the success rate of declining ron and

getting tsumo or declining tenpai to make a better hand as high? Usually not.

However, if we can make a guaranteed comeback hand, we should take it even if the wait is

worse, since the win rate of a bad wait is about 2/3 of a good wait.

Similar strategies can be used when calling. If calling would net us a conrmed comeback hand,

it’s ne to call aggressively even into atozuke or kata-agari hands. In all last, players will often

push everything which makes calling especially eective.

醍醐味 daigomi (n): 1. the real pleasure (of something); the real thrill; the true charm

2. avour of ghee; delicious taste

3. Buddha’s gracious teachings (Buddhist term)

150 CHAPTER 4. POINT SITUATION JUDGMENT

We can call even with 1 han less than needed and hope to add an aka or omote dora, create

another yaku or call kan. If we think the hand would be dicult to complete closed, we should

make calls like these.

Declining wins in all last

A situation that often comes up is when we call riichi in all last, and our winning tile comes out

from the wrong opponent, so that we’d need ura dora to come back if we call ron, but will certainly

come back if we can tsumo.

The chance of getting ura dora is about 30% (depending on the number of unique tiles in the

hand). If we don’t care about losing placement (currently 4th), we should decline if the chance of

getting tsumo is higher.

Of course, this depends heavily on the turn and on how many of our winning tiles are left. Also,

we must take into account that an opponent might win before we can tsumo. We therefore need

that

(1 − opponent win rate) × tsumo rate > ura dora rate

We’ll assume that East and South get 18 draws, while West and North get 17, so we should add

1 turn to the values given here if in West or North.

With a bad wait, the tsumo rate with 1 tile discarded is only 36.8% even on the 1st turn. Taking

about 10% for opponent win rate, we should always declare ron with a bad wait.

With a ryanmen, the calculation is more dicult. Assuming all opponents fold, it’s good to

decline with 7 tiles and 4 draws left or 5 tiles and 5 draws left. However, if an opponent cut our

winning tile, it’s safe to assume at least one opponent is attacking. If we assume one opponent is in

tenpai (with a good wait 2/3 of the time) and the other two fold, we should only decline if we have

at least 12 draws left. If we assume the folding opponents will deal in 10% of the time, we should

always call ron. If we assume the attacking opponent is not yet in tenpai, it’s good to decline with

a good wait until about the 13th turn.

However, we have disregarded the possibility of losing placement or a renchan completely, so in

practice deciding will be more dicult, but we can use the same kind of reasoning, knowing that

the chance of ura dora is roughly 30%.

Speed maximization techniques

Often, we won’t care about score in all last (when we get 1st with any win).

When this is the case, we need to build our hand to maximize speed. Dora and its neighbors

suddenly become worse than regular tiles since they come out harder. All yaku after the highly

important rst one become unnecessary. However, going for no yaku riichi is not a problem either.

We’ll have to make plays that maximize win rate to the utmost.

Example qqetuidgjkkcbmdora q→ cut qand go for tanyao

Since we want to maximize win rate, we should go dama if we have a yaku. However, we

should take a good wait no yaku riichi over a bad shape dama.

Example rtyyyzzxvbnm,. → cut x and riichi

Example tyyyzxccvbasd4→ cut 4 and riichi!

151

However, in the lower example, if calling riichi would lower our placement, we should stay dama.

Other useful plays are those that upgrade the shape, ignoring score

Example eerrty→ cut e to make sanmenchan with u over iipeikou

Example wrtyudfgggbnm→ call any of tyisg

Example uuiiaaas

7→ chii yo

However, when playing with rulesets where raw points or shuugi are important, we don’t need

to maximize speed as much.

When to attack fast

There are other situations where a swift attack is useful, such as

• Slim point dierence both upwards and downwards

• Improving placement by 1 is easy, but improving by 2 is unrealistic

• Improving placement is unrealistic, but there’s a slim dierence downwards

• We want to avoid yakitori

Often, we should emphasize preserving our placement over improving it when being overtaken

is likely. While people often say “in mahjong, take 1st at all cost” and there is some truth in this

(rulesets where 1st place gets a big reward are common), it’s no use to chance an impossible 1st

place.

With standard 45/5/-15/-35 placement points, players who get 30%/20%/20%/30% (average

rank 2.5) results and those who get 26%/26%/24%/24% (average rank 2.46) will make the same

amount of money. It’s both much easier and much more eective to improve average placement by

0.05 than to improve top rate by 5% (this is extremely hard). To raise expected placement,

rather than chase many 1sts, it’s much more eective to slightly increase average

placement by preserving placement when going for 1st is unreasonable.

Calling when a comeback is uncertain

When we get 1st with any win, we should of course call from anywhere maximizing speed. Then

what about the following situation?

If we call, we get an uncertain comeback (needing to tsumo or win with a particular tile), if we

stay closed or calls something else, we can make a comeback with a win from anywhere.

The situation is similar to the theory of waiting for an upgrade we discussed earlier (a hand we

can only win under certain conditions is like a bad wait or a furiten). Accordingly, if no opponent

is yet in tenpai, we should decline the call if we can draw at least 7 kinds of tiles that

would give us a conrmed comeback hand. Of course, if an opponent is already tenpai, we

should take even a bad tenpai to catch up.

Problem 6th turn, dora 4, 1300 point dierence with 1st place (non-dealer)

wwrrpxc

,. 7

152 CHAPTER 4. POINT SITUATION JUDGMENT

If we draw or call pon on wr, we can only get 1st with tsumo, but with any of eyzv

tsumo or chii (so 8 types), we can win from anywhere. Accordingly, we should not take tenpai with

wr.

Push-fold judgment in all last

Placement points of course have a big inuence on push-fold judgment.

Suppose that under standard rules (25000 start 30000 end, 10-30 uma, so 45/5/-15/-35), we’re

in 2nd while an opponent has called riichi, and that if we win, we will certainly be rst, but if we

deal in, we’ll certainly be last. When should we push in this situation?

Since both of these scenarios represent a change of 40000 placement points, we should push if

win rate > deal-in rate. Even if they’re the same, we should push since the opponent might tsumo

and lower our placement. Since this will be the case even with a bad wait tenpai, we should attack

with all but the most dangerous tiles.

The decision gets more dicult in iishanten. With a wide iishanten (about 20% chance to draw

into tenpai), we should push, but with a narrow iishanten (about 10%) it’s no problem to fold.

The probability of an opponent tsumo or opponents dealing into each other is dicult to estimate

making push-fold judgment in all last very complex. We can gure out which opponents will push

or fold not just from their discards, but also from their placement and point dierence (last place

will push, an opponent who can’t realistically lose placement will also push, 1st place with a big lead

will often fold). In general, we should start by thinking how much point each placement is worth,

then estimating the chances and eects on placement of winning, dealing in, opponents dealing into

each other and getting tsumo, and weighing the relative expected placement points of pushing and

folding from there. Since it takes a lot of time to estimate all of these variables, we should start

thinking about this as soon as all last starts.

2nd place, winning gets us 1st, dealing in gets us 3rd but certainly not last

We should attack even if win rate/deal-in rate = 1/2, so with any iishanten or even a good 2-shanten.

Conversely, if we can’t get 1st and can get 4th, we shouldn’t attack unreasonably. Even if folding

would certainly get us 3rd, we should still only push if win rate > deal-in rate.

1st place in tenpai and an opponent is also in tenpai

If the opponent is a non-dealer and we can read that dealing in will let us keep 1st, we should

attack. If we are under threat of losing 1st place from the other 2 opponents and our hand is

slow to tenpai, we should even deal in on purpose (if our hand is fast it’s better to attack). While

reading the wait is hard, it’s not hard to discard tiles that are safe against the other two opponents

but not against the third. Even against riichi, it’s rare to deal into a baiman (or haneman when

playing without aka) so it’s not necessary to fold against a non-dealer if we’re safe against such

a deal-in and threatened by the dealer. However, if we’re playing in a system where raw points

impact results, we should almost never deal in on purpose.

If dealing in would drop us to 2nd, while a tsumo or ron of another opponent would let us keep

1st, we should usually fold even from tenpai. On the other hand, if we aren’t safe against a tsumo,

we should push unless just it’s before a draw and we’re safe against no-ten payments.

153

If dealing in would drop us to 3rd, we should push with a good wait until the mid game and

with a bad wait in the early game. If dealing in would drop us to 4th, we should only push with a

good wait in the early game and otherwise fold.

Push-fold judgment one hand before all last

Since our placement in South 4 is very important, thinking ahead to our possible placement there

will also inuence our push-fold judgment in South 3.

Since losing 2 placements in South 4 without dealing is quite rare, there is a strong correlation

between our placement after South 3 and South 4. Improving our placement in South 3 even by 1

place therefore has a big impact on nal placement.

However, we shouldn’t focus on “being rst at the start of all last” too much. If we have dama

3900 in South 3 and 1st place is ahead by 2000, we can indeed get 1st place with dama ron from

anywhere. However, we’ll only have a 1900 point lead which is very easy to lose, so we should riichi.

The same is true if we have a small lead in South 3.

Same as in South 4, it’s no problem to rely on luck to improve placement in South 3, especially

since we’ll get another chance in South 4.

It’s also often possible to win a cheap hand in South 3 that will narrow the gap to make it easier

to come back in South 4. For example, if we’re West in South 3 and trailing 10500 behind rst

place North, we need a haneman ron on someone else or haneman tsumo. However, if we win a

1000 point trash hand now, we can take 1st with just a mangan tsumo next round. Getting mangan

tsumo is much easier than haneman so this is quite important.

What has been said about all last can be used in South 3 too, but to a lesser degree. We

especially don’t need to be severe in attacking with bad hands. If we’re last place in South 3 and

our hand is fucked, it’s ne to sit the hand out.

How to make mangan

The most useful skill in all last is making the fastest possible hand, since 1/4 times we’ll be 1st

(even more if we count all the cases where we want to attack at maximum speed). The second most

useful skill is how to mangan, since situations where we’ll need a mnagan to improve placement are

common and mangan is often a realistic goal (unlike haneman and above).

However, if we’re 2nd and need a mangan to overtake 1st, while 3rd and 4rd can overtake us

easily, we should go for mangan only if we get lucky draws and attack quickly with a cheap hand

otherwise. Conversely, if the players below can’t overtake us even with a mangan tsumo, we can

go for a comeback hand. Even if we’re dealer, we should try to go for one big strike than hope to

chain renchan.

154 CHAPTER 4. POINT SITUATION JUDGMENT

Since we can’t get overtaken even with a mangan tsumo, we should cut 5and aim for mentanpin

(with sanshoku or aka if we get lucky for a guaranteed dama comeback). If the players below us

were close, we’d also cut 5, but this time we’d call anything and go for open tanyao.

The key to chasing mangan is dora. If we have dora 3, we only need 1 more yaku, so we should

call everything if we have one. If we have a dora pair and can call, it’s also usually no problem.

Since we can get a 3rd omote or aka dora or make another yaku, we’ll end up getting mangan

more often than when staying closed. If we have a single dora, we should hold on to it even if it’s

dicult to use. If we draw a second one, we can get mangan with just riichi and tsumo, or even

with riichi tsumo dora 1 ura 1 if we don’t. If we have no dora, it’s usually good to try to make a

closed yaku like tanyao or iipeikou (+ riichi tsumo ura 1) and hope to draw dora on the way. Of

course, we should be aware of sanshoku and ittsuu, but it’s not needed to break up groups to chase

them unreasonably.

However, there are hands where even reaching closed tenpai, not to mention winning a mangan,

is very dicult, for example if the hand is slow and has no dora or visible yaku (or the dora is

completely useless). In hands like these, we should accept that it’s unreasonable and go for a forced

honitsu (especially with an honor dora) or chinitsu in the suit we have most of.

We can especially

make aggressive calls like this if we’re unlikely to be overtaken.

The author seems to have forgotten about the shortcut to haneman. Chiitoi can be very powerful, but without

dora, requires riichi chiitoi + 2 (ura or tsumo + x) to make mangan, which is quite hard to pull o (remember that

chiitoi has a lower chance of hitting ura than other hands).

155

What to do when far behind

Hanemam needed for comeback

If we need at least haneman, we’re in for trouble. While aka help a bit, it’s still no easy task. The

most common and easiest to make haneman are mentanpin tsumo dora 2 and mentan/menpin

tsumo dora 3, so ecient closed hands. If we have few dora, we can think of stacking iipeikou,

sanshoku, ittsuu with pinfu, and if we still end up at mangan we can hope for ippatsu, ura

or a direct hit. The easiest open haneman are chinitsu + x, honitsu yakuhai dora 3, toitoi

yakuhai dora 3 and the like. If we have no other option, we can go for riichi tsumo chiitoitsu

dora 2 as a last resort. Since we’re likelier to get ura than draw dora if we don’t have any, we

should instantly riichi with chiitoi only if our wait looks drawable.

If we aren’t last place and need a haneman to improve placement, we should usually not chase

it unless our draws are excellent and focus on not getting overtaken. If we need to avoid all last

and need a baiman or more while our hand won’t allow it, we should try to secure no-ten payments

to narrow the gap or deal into the dealer for a renchan as a last resort.

Settling for last

If even a haneman tsumo isn’t enough, or our hand is too bad for this hand, and we are certain the

match will end after this round, what should we do? If playing in a placement system, we should

go for even the slimmest chance at a comeback. However, if raw score or other factors (shuugi,

yakitori etc.) play a role, it’s a dierent story.

If the chance of our placement changing is negligible (for example also if we’re rst place dealer

with at least a 30000 point lead), we should just maximize winnings from this round, so play just

like we would in East 1.

How to play when a player is close to shadowrealm

The hanchan can also end before all last if a player runs out of points. Therefore, if someone is

close to busting out, we should play with the same intuition as we would in South 4. If we’re close

to busting out ourselves, or 3rd when an opponent is close, we should attack unless we have a weak

hand and want to bet on a draw. If shadowrealming an opponent with dama would net us 1st and

calling riichi would make him fold, dama is eective. When deciding whether to decline a win from

someone else, we can think in the same way as we do in all last.

In mahjong, any lead no matter how big can be overtaken, so the advantage of ending the

hanchan is big. Even if we already have a lead or it’s still the East round, we should assertively try

to shadowrealm an opponent to end the match. For example, if haneman tsumo would shadowrealm

someone and let us overtake 1st place while we have a mangan tsumo dama, we should riichi even

with very few draws left. It’s no problem to decline a ron o the wrong player here. If we’re rst,

we’ll want to fold more, but if we have a shadowrealming hand we’ll want to counterattack (of

course without declining wins).

If a ceiling (hanchan ends when a player gets more than x points) is in play, the same ides apply.

Riichi sticks and honba

Sometimes we’ll play dierently because of riichi sticks and honba. Especially in rules where each

hinba is 1500 points their inuence is big. While winning fast becomes more eective with more

156 CHAPTER 4. POINT SITUATION JUDGMENT

sticks in play, it’s not that important to change our playstyle too much. For example, we should

still call riichi with good waits below mangan. In general, when there are many sticks, win rate

becomes more important.

Example With 3 riichi sticks, we can push against riichi even with 1000 good wait.

Example When unsure about whether to call riichi with a good wait mangan in the mid game,

we should stay dama if there are 2 riichi sticks.

When to call kan

Basics

The player most likely to reap the blessings of kan is the one closest to tenpai. Therefore, we should

call kan if we think we’re that player.

If there’s an opponent equally close to tenpai, we call can

kan only if the added points would benet us more.

When opponents are no-ten

• If we’re in tenpai. (If we have an open mangan where one more dora wouldn’t give us haneman,

it’s dicult. If the extra draw is important, call ankan and kakan. If exposing part of the

hand would make opponents too wary, lowering win rate, don’t ankan or daiminkan. If not

calling kakan would throw o opponents, we can decline it too. But if the gain from a dora

would be big, we can call kan even at the cost of scaring opponents.)

• Good shape 1-shanten in the early or mid game.

• Very good shape 2-shanten in the early game, for example rtyyaaaafjjccv

When an opponent is in open tenpai

• If we’re in riichi, kan.

• If we’re in open tenpai, kan when win rate > deal-in rate. Even with a bad shape, kan if the

point increase (3900→5200 etc.) is big.

• Don’t call kan in no-ten.

When an opponent has called riichi

• If we’re in riichi, don’t kan if we have a bad shape and the tile is safe. Otherwise kan.

• If we’re in open tenpai, call closed kan on a dangerous tile. Otherwise, don’t call kan unless

the point increase would be big (3900→5200 or 8000)

• Don’t call kan in no-ten.

Together with these basic rules, we should adjust a little for point standing (kan less when in

the lead and vice versa). However, it’s still ne to kan in tenpai when in the lead for the extra

draw. Conversely, we shouldn’t call kan if our hand is fucked even when we’re losing.

This is the old theory. The modern theory is to call kan whenever we feel our chance of winning the hand is

bigger than 1/4, for example with a callable 3-shanten in the early game.

157

Using kan components

When we decide not to call kan (yet), we should get of rid of the 4th tile if we don’t plan on using

it and we have a more useful tile. The benet of a faster hand is often bigger than that of kan.

Similarly, we shouldn’t call kan of doing so would hurt the hand.

Example qwetyujjjjcvm,→ don’t call kan

Cutting j guarantees a good shape tenpai with xvn..

Example wrtttpuilllcvbdora w → call kan

While we lose a good shape if we draw yo, we can go for e kanchan for 6400 or mangan

dora tanki, while if we didn’t call kan and cut w into ryanmen tenpai we’d only get 2600. So we

should kan.

Exceptions

We should also call kan when:

• The extra fu would give us a conrmed comeback in all last

• We can steal the haitei from an opponent in riichi

• We’re last in all last and don’t care about losing points

• When an opponent dealing into a big dick hand or the dealer getting hit by a big tsumo would

improve our placement in all last

完