McDowell, David - Blackjack Ace Prediction. The Art of Advanced Location Strategies

- BLACKJACK
ACE PREDICTION
DAvID McDowELL
: SPUR OF THE MOMENT PUBLISHING
Digitized by the Internet Archive
in 2021 with funding from
Kahle/Austin Foundation
https://archive.org/details/blackjackacepred000O0mcdo
BLACKJACK ACE PREDICTION
The art of advanced location strategies for the
casino game of twenty-one
For Flora
BLACKJACK
ACE PREDICTION
The art of advanced location
strategies for the casino game of
twenty-one
DAVID MCDOWELL
With a foreword by Michael Dalton

Copyright © 2004 by Spur of the Moment Publishing
First edition / First printing
~ Printed in the United States of America
09 08 07 06 05 04 OSS]. Guat See) 1
ISBN 1-879712-10-5
Library of Congress Control Number: 2004110090
McDowell, David.
Blackjack Ace Prediction: The art of advanced location strategies for
the casino game of twenty-one.
Includes appendices, bibliographical references and index.
All rights reserved. No part of this publication may be

translated, reproduced, or transmitted in any form or by any
means, electronic or mechanical, including photocopying and
recording, or by any information storage and retrieval system,
without the express written permission of the copyright owner.
Inquiries should be addressed to Spur of the Moment
Publishing, P.O. BOX 541967, Merritt Island, FL 32954-1967
USA.
Visit our web site at www.BJRnet.com to order copies of this

book and for possible updates, errata and to participate in our
message forums. For wholesale and retail information email
editor@BJRnet.com or contact us at the address above.
Blackjack Review Network

Gaining an edge in cards and lite!
PUBLISHER / EDITOR:
Michael Dalton, Spur othe Moment Publishing, Merritt Island,

Florida,
editor@BJRnet.com |www.BJRnet.com.
COVER ARTWORK CREDIT:
Copyright © Directional Publishing, Inc., “Ace of Spades” by Abigail T.

Kamelhair.
PHOTOGRAPHIC CREDITS:
Edward Thorp, courtesy of New Mexico State University Library Archives,

Las Cruces.
William Earl Walden, courtesy of Washington State University Library

Archives, Pullman.
Edward Thorp at University of California, Irvine, from I Have a

Photographic Memory by Paul Richard Halmos, courtesy of the American
Mathematical Society, Providence, Rhode Island.
Shoulder holster, courtesy of Warren of California, San Diego,

www. travelsecurityaccessories.com, 800-932-4465.
DISCLAIMER:
Although the author and publisher have exhaustively

researched all sources to ensure the accuracy and
completeness of the information contained in this book, we
assume no responsibility for errors, inaccuracies, omissions or
any other inconsistency herein. Any slights against people or
organizations are unintentional.
Gambling infers risk. The author and publisher accept no

legal responsibility or financial liability whatsoever for any
reader’s application of the information contained in this book.
TABLE OF CONTENTS
Tables and Figures icc.::-sscsaceesesranscese.

cacseeqnrt Waccnacnece-eeesueecerd 10
A CKNOWICASINENES: jc ccc.cc:ssaesnncecns
sebccecsasdacarsenacecereecsevanrdiunes 12
About the Cover: Art sasiizi.scssstevsaceechsaceutsvscescsvetcacweseereneeds 14
Foreword by Michael Dalton .............c.sscsccssscssessesccenscenees Lbs
Chapter] =A. ce Prediction i... s1ecc hess nae coe ee 17
Doing ShufflesAnalyses- Seine

ee ee ee ee ee 19
Dome Computer Simmlcrians 2, eee
eee ee 20
Usine ihe Hx peciation Honmulge ee. . ee ee cn |
Other’ BOOKS: sist sci, Reena the kolo 22
Published Referencesmet.. so. a ee ee 23
“How I Invented Ace Prediction ””—Ed Thorp ..............-.- 30
Table of Contents
Chapter 2 - Non-Random Shuffling ........ccccccscscsscsceeseeeees 32
Permutations by Cutting and Shuffling .....cccccccccccc0c0000000+. aL

Thorp’s Non-Random Shuffling Analysis.........0cc0c00000000-+. 34
Epstein’s Analysis of the Imperfect Shuffle..........c0ccc000. a7
Casino Card Shuffles: How Random Are They? .............. 39
inpormation Loss Card Shujpline sates ena 40)
Babiana; Mystery: Charles T. JOP vitcstsscpsu concesaseese: 43
Chapter 3 - Analyzing Shuffles..........cccssssssecssscesssoveseareres 45
DPIVINg NOT pS SHUI ANOIVSISU see eps ee 45

|ue .
ELORCTE SCQUCHGES < MErn Sate. ues Seen ee 54
OUNMNS INE BICAKSocnencosss0 ee Rise ee Be ee 2s)
EMOUADINILY Of BI OBEN, SEQUCNCES 00a eee ect ee 59
DraKenisequences Dy. COMPULEIiit. teenie
eee cee 60
PECLNOL EeIUGS, SITIOS AGB OX MOUS wens. rete ee 61
PATI TLEVID LO LOT OMG ch etait tia cate fons oases eee ee ate eee 64
Chapter 4 - Exploiting Shufflles ............ssccssssscsssssereesscesees 66
Applying Hannum ’s Shuffling ANALYSIS .......2c00sc0evereere 66

SIC MleSIRO TEINGNAOMMES Sa 2 scapes -opsondecwi
sesrarer
p 69
VACHS SILI CW) COIICSSCS Sc cacoar encerersseeec ace 72
cer car
estnee
HoweReliable Are the Weaknesses? t.nccusss.nvsosasevacsenncnn/242- 74
Prearcine Exel POsitions Of CATAS 0. sse.-oh-ticecns-ncnseosntte 76
MCN OTN OICI IU ite voce pve alae naan dace rads oieh sere danevegetr ana si
Table of Contents
Chapter 5 - Locating ACES ........sssseseesesssssrsrenseerencensesensnees 79
ACE LOCALION Wsicucsc a eee eric nee, eae eee 79
ACE SIUGS «.ncncnnsnnsect tis Does tay settee e ten Met dae Oana ates ee era 80
A Stepladder Shuffl ei a eve cetacean eee &3
4 Two-Pass ComboiShufflese. ee eres eee 85

The. Greek Syndicate:.:, eee cee ne eee 8&7
Chapter 6 - Predicting ACES ..........:cssscssssseerecerssserssenesseees 89
Card PeGQiCtiOn caro ea ee coe eee 8&9

The Prediction Problemac ae ea ee eee 90
Applying Epstein 's Prediction Theorem..w255-..-veae= 91
ine Prediction Problem S0ived. 122.2500 94
A Prediction Rule nmin RRR ee 96
How to. Memorize-Key Cards) 96
Memorizing the Target Ace. pom. a eee ee ee 97
ISLCETING LCA CES se cnscirastastaa
en eetancnniy 99
ECUSCA CN. C CICS en rss Stir
a 6 eee 100
POULET CAPASSO aA nD RT tO a 102
Putting ICANT O@eihery. «ance ee ete ee ee 104
Te: GUrse: OF WONG a. cexacscur, oes eee Genre ete nee 106
8 Table of Contents
BEAL Ne, DCCC yAram SEWN AL Fo cice en cfoRedvniheteeiewcsien 108
VOLUesG) MACOS FIrst CAN, 26.31 ea 108
How Often Does the Ace “Hit the Money”? .......cc00.000.:. Tt
Skilled versus Unskilled Predictors ...ccccccccccccccscccsssssesees Li
When ine Wealer Gets the ACen 0 es
I QTIICMNOLICOL EX PECIAIION., See oe. Re ci cosensdsssecsceesss, 113
PLeOIClaDULY Of.COSINO-SRUJIES. a. a eee etek: TTS.
ESC OUNE OPC TGIID UNIO: «0.2 A, ts SRR a by p08 118
Chapter 8 - How Much to Bet.............cccssssssssssssssscsssssece 120
CCIE RCH Hates ter dene Ns cae 2 Wig.Cees. SLR aR 120

JPRS oo ob, reel 2 BRE ea ko a ORT mR 122
ODUMIGL DCE SIZ Cre: cs On AtOR h Atte LTR Aenea 126
DOIN
SP GL ORGCEMENA 5k. er ee ec eh aaah ite aes NAT:
AD DCTUICES ¢cecessssorsecoceduessstncccesesoseonassoesvassausestcaase<sentesesss 130
PADDCHOIN A. SIM JUNG 1)

CYININO|OOV = an. peteede 130
Appendix B: Card Movements in Perfect Shuffles.......... 131
PATCH GIN AM ANICOTN OL Up etme ieiete eho tke ade eerie: 133
ADD CNA DAD EY LESTE DBRS Fob. det Ree war a 135
Appendix E: Kendall’s Concordance Coefficient.......... 136
PAD DCH GQIISCL CST te. stcpensessendoe.cebonssteoees NESE
FUSCA =W
NUON soca scree aeta natn oe ee ac ccnassecceresecaedeaccs-ciccsencesieccsaaets 139
RGLEF ENCES 2. .50.20csceenscsveccnsccescesscstssaccsssanassdsansesneseseseroutess 146
Table of Contents 9
TABLES AND FIGURES
(DABLE2-1., A-PERFECT SHUEFLLE c.c.cc-cccsccesesseeccnecesessececececceovestess 33

TABLE 2-2. A MODIFIED PERFECT SHUFFLE. .......seccsssesecececceereees 33
TABLE 2-35 CARD PERMUTATIONS :-cseec.cs-ecccees-cecetecteecertessnesssseeese 36
TABLE 2-4. CARD TRANSITION PROBABILITIES .......cccccceseseesessees 36
TABLE 2-5. CARD PACKETS OF VARIOUS LENGTHS ........cccsseeesees 39
FIGURE 2-1. CARD PACKETS OF VARIOUS LENGTHG........+++22020022 39
TABLE 2-6. HANNUM’S SHUFFLE RESULTG.........ccessssssssssccesessseees 40
FIGURE 2-2. HANNUM’S SHUFFLE RESULTS ...........cccccsssssssceeeesees 40
TABLE 2-7. INFORMATION LOSS IN CARD SHUFELING ........00000008 42
FIGURE 2-3. INFORMATION LOSS IN CARD SHUFFLING.......s0ss00e 42
TABLE 3-1. CARD GAPS FOR A ONE-RIFFLE SHUFELE .........00000 47
FIGURE 3-1. CARD GAPS FOR A ONE-RIFFLE SHUFFLE.......0000000 47
TABLE 3-2. CARD GAPS FOR A TWO-RIFFLE SHUFFLE........000000 50
FIGURE 3-2. CARD GAPS FOR A TWO-RIFFLE SHUFFLE........--+- 50
TABLE 3-3. CARD GAPS FOR A THREE-RIFFLE SHUFFLE ..........- 52
FIGURE 3-3. CARD GAPS FOR A THREE-RIFFLE SHUFELE.........- a2
TABLE 3-4, PROBABILITY OF A BROKEN SEQUENCE ....:esceeceeeeees 59
TABLE 3-5. PROBABILITY OF BROKEN SEQUENCES. ......:esceeseeseees 60
FIGURE 3-4. PROBABILITY OF BROKEN SEQUENCES .....cecceeseeeeeee 60
TABLE 3-6. BROKEN SEQUENCES BY COMPUTER .wescccceccesseceeeseeee 61
TABLE 3-7. EFFECT OF PLUGS, STRIPS AND BOX CUTS....ccceseesees 63
TABLE 3-8. A BROKEN SEQUENCE ADJUSTED ..ccscececcececcesesseccesees 63
TABLE 3-9. BROKEN SEQUENCES ADJUSTED ....ccsecccesscssccessecceecese 63
FIGURE 3-5. BROKEN SEQUENCES ADJUSTED.....eccccscccesesssessceceeee 63
TABLE 4-1. THE AUTHOR’S SHUFFLE SIGNATURE cccessccccoseosesecece 66
FIGURE 4-1. THE AUTHOR’S SHUFFLE SIGNATURE cecccscccosececeeceee 66
FIGURE 4-2. PRE- VS. POST-SHUFFLE PLOTS ccscscccccccsosecesocoseseseee 68
TABLE 4-2. PRE- VS. POST-SHUFFLE CARD POSITIONS ce... 69
TABLE 4-3. FOUR SHUFFLES BY HAND (16) orscoreccsccontcteenccemacetcees
70
TABLE 4-4, FOUR THOUSAND SHUFFLES BY COMPUTER eeesess...-..
70
FIGURE 4-3. IDENTIFYING SHUFFLE WEAKNESSES \icccecesccsevececece
Yo?
FIGURE 4-4. MAPPING SHUFFLE WEAKNESSES
10 Tables and Figures

TABLE 4-5. A SIMPLE REGRESSION ANALYSIS ccccccscoceececceccecceseece 74
FIGURE 4-5. THE LINBO BEST UD <.sco.cosecsectsotesdcecsecosicivececcseiecc, 74
TABLE 4-6. PREDICTION INTERVALS .......cccccsssssscccscccocceosseceeceeeee day
FIGURE 4-64 PREDICTION BANDS <sccsocenssasessonssocssczevesecacessoososooeeess dal
TABLE 5-1. MOVEMENTS OF A SINGLE KEY CARD vecccocccscccceceeeee 79
RABULES—2. CROBABIONUY OF (ACES osiscti recsceescce
ccivedensnopisarcseiens
s 81
FIGURE Solo PROBABILIEYI OF A CES ii:.5..cct.cnccseeeckendee ee. 81
TABLE 5-3. AT LEAST TWO CONSECUTIVE ACES.....cccceccsscosssesees 82
FIGURE 5-2. AT LEAST TWO CONSECUTIVE ACES. ....cccccccoscseceoeee 82
FIGURE 6-1. A SIMPLE FOUR-CARD DECK. .......cccssceseccessccescccseces 90
FIGURE 6-2. THE PROBLEM OF PREDICTION ......cccccsssccesseccesscceeee 90
FIGURE 6-3. MEMORIZING THE KEY CARD .......ccccsssssccsssscecsssceees 92
FIGURE 6-4. A SEQUENCE IN ITS INITIAL ORDER ......ccccecccceseeeeee 92
FIGURE 6-5. THE KEY CARD IN THE FINAL ORDER........ceceeeeseeee 93
FIGURE 6-6. THE KEY CARD RE-EMERGEG........ccccccssescsccccccssssces 93
TABLE 6-1. THE PATTERNS OF PREDICTION. .......ccccccccsscssssceecseees 94
FIGURE 6-7. THE TARGET ACE APPEARS ......cccccccccceccessessssessessees 95
TABLE 6-2. ADDITIONAL ACES FOR SIX DECKS ..........00000000e00000s 98
FIGURE 6-8. ADDITIONAL ACES FOR SIX DECKS .........ssesccsseseeeess 98
HIGHER 6-9. SPP ERING THEVA GES coccorcccssocscaccnssessseeccsercscesceecesece 100
TABLE 6-3. PROBABILITY OF A FALSE KEY CARD..........0cc0ee00e8 101
FIGURE 6-10. PROBABILITY OF A FALSE KEY CARD. .........cceeee 101
TABLE 6-4. PROBABILITY OF FALSE KEY CARDG..........00eseeeeeees 102
FIGURE 6-11. PROBABILITY OF FALSE KEY CARDS .........eseeeeeee 102
FIGURE 6-12. ASYMMETRIC FACE DESIGNS. .........ccccccosossesssseees 102
FIGURE 6-13. ASYMMETRIC WHITE SPACE. ........ccccccccseecessenseees 103
FIGURE 6-14. ASYMMETRIC BACK DESIGN ..........scccccccosssssseseees 103
FIGURE 6-15. “READING” THE EDGES OF A BEE® DECK ........ 104
FIGURE 6-16. AN ACE IN A TRACKABLE POSITION...........ceceeee 105
FIGURE 6-17. A FALSE KEY CARD APPEARG..........cccccccssessssseeee 106
FIGURE 6-18. THE ACE FALLS ON THE PREDICTED BOx.......... 106
TABLE 7-1. EXPECTATION WITH ACE AS FIRST CARD..........0066 109
TABLE 7-2. EXPECTATION FOR ALL FIRST CARDS......sssccceseeeeee 109
TABLE 7-3. SKILLED VS. UNSKILLED PREDICTORS. ..........seeeeee0 1
FIGURE 7-1. SKILLED VS. UNSKILLED PREDICTORG..........s0e0000s 112
TABLE 7-4. VARIANCE FOR THE EXPOSED ACE .......ccsesessssseeeees 117
TABLE 8-1. NUMBER OF BETS PER HOUR ......ccsccsssesesesessecsseeseees 122
TABLE 8-2. RELATIVE BANKROLL GROWTH RATE.........0esseee000 123
TABLE 8-3. DOUBLING BEFORE HALVING.........ccccccceererreesseeeeees 124
TABLE 8-42 BVER WOSING TAU Besccicccsssssscoccscsacccseocecsesssssssensensees 125
FIGURE 8-1. OPTIMAL FRACTIONS OF A KELLY BET .........000+0 125
Tables and Figures 11

ACKNOWLEDGMENTS
Thanks to:
“Ace”, www.ace-ten.com
Michael Barnett, East Perth, Western Australia
Dr. David Bayer, Columbia University, New York

Dr. Rodney Carr, Deakin University, Victoria, Australia
Charlie Choo, Alakazam! magic shop, Singapore
Anthony Curtis, Las Vegas Advisor, Las Vegas, Nevada
Dennis Daily, New Mexico State University, Las Cruces
Michael Dalton, Blackjack Review Network, www.BJRnet.com
Dr. Persi Diaconis, Stanford University, California
Dr. William R. Eadington, University of Nevada, Reno

Dr. Edgar N. Gilbert, Bell Laboratories, New Jersey
Dr. Solomon W. Golomb, Univ. of Southern California, Los Angeles
Dr. Robert C. Hannum, University of Denver, Colorado

Dr. Eliot Jacobson, University of California, Santa Barbara
Susan Jarvis, University of Nevada, Las Vegas
12 Acknowledgments
Martin Joyal, Pierrefonds, Quebec, Canada
Abigail T. Kamelhair, Atlanta, Georgia
John. K. King, Detroit, Michigan
“Marko,” The Learned Pig Magic eZine, Panama

Anthony Maxwell, The Heraldry Society of Scotland, Edinburgh
John May, Dublin, Ireland
Rob McGarvey, Toronto, Canada
David Morse, Cranston, Rhode Island
David Nichols, Birmingham, Alabama
Darren Phillips, New Mexico State University, Las Cruces
Michael Porstmann, Dachau, Germany

Richard Reid, www.bjmath.com
Donald Schlesinger, Bardonia, New York
David G. Schwartz, University of Nevada, Las Vegas

Howard Schwartz, Las Vegas, Nevada
Frank Scoblete, Chicago, Illinois
Ralph D. Slocum, Warren of California, San Diego

Arnold Snyder, Las Vegas, Nevada
Stanley R. Sludikoff, Inglewood, California
Suzy Menkes Spanier, Neuilly Cedex, France
Lawrence R. Stark, Washington State University, Pullman
Dr. Edward O. Thorp, Harbor Ridge, Newport Beach, California
Norman Wattenberger, www.qfit.com, New York

Victor Wise, www.cardsorcerer.com, Los Angeles, California
Stanford Wong, La Jolla, California
Bill Zender, www./astresortconsulting.com, Las Vegas, Nevada
Acknowledgments
ABOUT THE COVER ART
Abigail Kamelhair taught art for 10 years before

becoming an independent artist in 1987. Her
charismatic Ace of Spades is one in a classic
series of four medieval playing cards, the others
being King of Clubs, Queen of Hearts and Jack
of Diamonds.
Queen of Hearts Jack of Diamonds
© Directional Publishing, Inc., 2004 www.directionalart.co

m
14 About the Cover Art

FOREWORD BY MICHAEL DALTON
The material you are about to read presents the end of an

incredible journey for the author and, I must say, for myself as
well. Gaining an advantage in the game of blackjack has been
an exciting challenge and profitable adventure for me since the
early 1980s. In the early years of my card counting
experience, I knew nothing about the possibilities of shuffle
tracking or Ace location strategies. Indeed, they were secrets
that awaited my discovery. I was finally introduced to shuffle
tracking concepts in the early 1990s and successfully used the
technique in the casinos. Only recently have I been privy to
the concepts behind Ace prediction.
In August of 2003, David McDowell emailed me a copy of
a report he had written. He titled it “The Poolside Guide To
Blackjack Ace Tracking’—a very cute title for one of the
most powerful winning techniques known to the inner-circle
of professional blackjack players. After giving it some serious
thought, I agreed to publish David’s work—which I thought
was the best material on the subject available anywhere. Well,
several months and a few hundred emails later David’s small
report had transformed into, what I believe is, the most
authoritative book on the subject.
There is no doubt that the basic Ace location strategy
discussed in the book was the “secret” of many of the
successful blackjack teams of the 80s and 90s. As such, there
Foreword by Michael Dalton 15

will be many players that will be unhappy that this
information has finally been published in detail. But I believe
the time has finally come to let the world in on the big secret.
You see, one of the big secrets to winning at blackjack
always had to do with the Ace card. We all know that the
Ace is the most important card in blackjack. Without Aces
you can’t get a natural 21 and getting paid three to two in
blackjack is where most of our profit comes from. But the
problem has always been in how to find those precious pearls.
That is where Ace prediction and location strategies come in
and that is what this book is all about. Are there any other real
secrets to the game? Well, I highly doubt it. But secrets are
such an elusive animal and | have been surprised before.
According to research I performed for my Encyclopedia of
Casino Twenty-One, there are over 200 books that have been
published on the game of blackjack, not to mention over 150
technical reports on the subject. But to be perfectly honest,
there are only a handful of books that are worth your time and
effort to study. In my opinion, this is definitely one of those
books.
Many will consider David’s work as the final chapter in
advantageous blackjack play—as the cat has finally been let
out of the bag. So pour yourself a cup of strong coffee, sit
back, and enjoy the journey.
Michael Dalton
www.BJRnet.com
16 Foreword by Michael Dalton

ACE PREDICTION
The winning blackjack technique known as Ace Prediction,

Ace Location, Ace Tracking, Ace Sequencing, or Key
Carding, can be extremely profitable given the right
conditions. In games with simple shuffles” it is capable of
delivering huge advantages over the casino. Such shuffles still
can be found, even in Las Vegas, but in recent years the
widely publicized coups of high-stakes professional blackjack
teams—such as the famous MIT and Hyland operations—have
led to increasingly complex shuffles. Many dealers now riffle
the cards three or more times, which makes Ace Prediction
much less profitable and, in some cases, unprofitable.
Sequence trackers will find in this book my formula for
identifying the profitable games.
For the first time there is a lot of detail about methods for
working out if a shuffle is predictable before you play it—and
how much you can expect to earn. The effect may be for more
players to work out which games are vulnerable, and there is
no doubt Ace Prediction can seriously hurt those games, but
analyzing shuffle predictability is typically the type of thing
* See Appendix A for definitions of shuffling terminology.
Ace Prediction 17
big money pros do that non-pros don’t do, so it depends on
how serious you are about winning.
The same applies to finding good shuffles. Blackjack as a
blood sport is something of a mature animal, and although it is
still surprisingly easy to find trackable games, you have to
travel farther to find them. Not a problem for professionals,
but it can kill the dream for the amateur. Trying to Ace track a
poor shuffle just because it is close to where you live, for
example, is amateurish.
Predicting Aces is a very simple idea and it’s easy to do at
the table. However, making it pay depends heavily on shuffle
analysis and selecting a profitable game. The difference
between profitable and unprofitable can be a few cuts or
riffles, so it’s very subtle and care is required. This book
provides the tools you need to find out if there is any edge
with a particular shuffle and, if so, how much, no matter how
complex the shuffle may look.
Some people assume because they are predicting Aces the
advantage will be huge. Others assume the advantage needs to
be huge. Neither is true. The advantage can be huge, but it
may also be the same ascard counting, better than card
counting, or there may be no advantage at all. A lot of games
are trackable to some limited extent, but many of those games
are simply not worth tracking.
If you get it right, Ace Prediction in modern games will get
you (conservatively) a 3-4% edge. There was a time, many
moons ago, when you easily could get double that—and
more! If you are diligent about scouting out predictable
shuffles, you can still get very high advantages, but anything
above 3% is an enormous advantage and the problem
becomes: how do you hide the fact you are winning?
When you have a big edge and are making very big bets,
the most important thing is how many hours you play before
they kick you out. The trick is to get enormous bets on to the
tables without the casino suspecting anything. Professional
teams use the big player ploy. One or two team members sit at
the table making minimum bets and tracking Aces, while the
“big player” hovers around and jumps in to make big
bets
18 Ace Prediction
when a secret signal is given that an Ace is due to appear. As
a result, casino surveillance staff dubbed the big players
“buzzards.”
To pull this off requires nerve, but getting big money on to
the table when you have the advantage is the real key—not the
size of your edge. Four bets at 2% is just as good as two bets
at 4%, and it looks less suspicious.
DOING SHUFFLE ANALYSES
The best work on non-randomness in human shuffling is

Bob Hannum’s paper “Casino Card Shuffles: How Random
are They?” (2000).'’ If you want to know how to analyze
shuffles using statistics you should study this paper.
For help in understanding Hannum’s paper, and this book
in general, I suggest Douglas Downing and Jeffrey Clark’s
book Forgotten Statistics.” Their probability chapter uses card
examples, and the book includes a 65-page “dictionary” of
statistical concepts.” A good book for complete beginners is
Learning Statistics Through Playing Cards,’ by Thomas R.
Knapp.
If you already know statistics, and want to delve more
deeply into the subject, the classic “cookbook” is Sidney
Siegel’s Nonparametric Statistics for the Behavioral
Sciences,’ which contains step-by-step instructions for doing
all of the popular nonparametric tests of statistical
significance, including the ones used in this book.
My shuffle study, like Hannum’s, measured how many
cards are interlaced between other cards during hand-executed
shuffles. My results agreed substantially with Hannum’s. The
single-riffle result was: (Hannum’s figures in brackets) 1-card
83% (77%), 2-cards 14% (18%), 3-cards 2% (4%) and 4-or-
more cards 1% (1%). The two sets of figures are in close
agreement. I also measured gaps between initially adjacent
* Footnotes using the notation /F'S/ indicate page numbers in this dictionary.
Ace Prediction 19
cards. For one riffle the mean gap was 0.97 cards (0.93), for
two riffles 2.9 cards (2.8) and for three riffles 6.7 cards (6.4).
The percentage of single cards interlaced is useful to know
when deciding how predictable a particular dealer’s shuffle is.
The number of cards between initially adjacent cards is used
when predicting Aces during the game—you simply count
how many cards follow the “key card” that tells you when an
Ace is due.
As you will see later, a number of statistical tests were
done which used the data generated by the hand shuffles. The
x test for randomness, and Kendall’s Concordance Coefficient
for the level of similarity between shuffles, can tell you which
shuffles are predictable and which are not. The x test is
particularly useful, because it exposes every tiny detail of
where cards tend to remain together and undisturbed through
the shuffle—weaknesses a player can exploit to predict where
Aces will appear.
DOING COMPUTER SIMULATIONS
The important statistical results in this book were verified

using computer simulations. Norman Wattenberger’s
excellent Casino Vérité® (CV) Shuffle was used for these
simulations.
The software provided confirmation of the probability of
intact and broken sequences in a shuffle. Instead of the
tedious process of having to count all the breaks in the shuffle
and then calculating all the separate chances of a key card/Ace
sequence being broken, the player can simply plug the
computer’s result into my simple formula and get the
probability instantly. CVShuffle also identified the exact same
two shuffle weaknesses revealed using pencil and paper
methods—in about one hundredth of the time!
The computer analysis was very slightly complicated by
the fact that, in my desk-calculator work, card positions were
counted from the bottom of the final stack upwards, whereas
CVShuffle counts card positions from the top of the final stack
downwards. \ used this counting method because it’s closer to
20 Ace Prediction
the way a player views the cards in the discard tray during
actual play. It is also similar to the shuffle maps used by
Arnold Snyder in his Shuffle Tracker’s Cookbook’ (segments
labeled A, B, C, D, etc., from the bottom up). CV Shuffle does
it the other way round to make it easier to build a model of a
shuffle. Of course, it doesn’t make any difference to the
statistical outcome. The results of the four shuffles I did by
hand were reproduced by the four thousand shuffles done by
CVShuffle in less than ten seconds!
Regarding “keying” the Aces: the BJAP method uses one
key card only. Conventional card-tracking wisdom says:
“always use three key cards” but, as Snyder points out
(see page 76 of his ST Cookbook), you can make money
using one key card as long as you are pinpointing Aces to
within half a deck.
Norm Wattenberger re-wrote the code of CVShuffle’s “Ace
Sequencing” analysis module to include one key card, as well
as one or two out of two key cards, so I could conduct the
simulations for this book. Two of the three shuffles in
BJAP—the two-pass combo and the one-pass_riffle-and-
restack—have now been added to CV Shuffle.
I also simulated a third shuffle, a two-riffle R&R that had
lots of plugs, strips, and box cuts. The higher level of broken
sequences completely wiped out the advantage of the Ace
tracker! This provides a stark warning to Ace trackers to
avoid at all costs complex two- and three-pass shuffles
liberally interrupted by plugs, strips and box cuts.
USING THE EXPECTATION FORMULA
Let me say a few cautionary words about the expectation

formula in Chapter 7. The most important decision an Ace
tracker makes is which game to play, and which to leave
alone. You must get that decision right or your bankroll will
suffer. The formula will help, but it wi// only be as good as
the accuracy of the figures plugged into it. For example, if
your estimate of the expected “hit rate” for catching Aces is
Ace Prediction 21
poor, the expectation figure will be “way off’ at best and
“downright dangerous” at worst.
It is very important to measure hit rate in the casino before
you play for big money. This is best done using the statistical
“clocking” method described in Chapter 7. Modeling a
shuffle on computer should be used only as final confirmation
of the empirical result obtained by monitoring the real shuffle
in the casino, not as your sole method for estimating the hit
rate of a shuffle.
OTHER BOOKS
Good information about Ace Prediction is scarce. The best

previous work appeared in Arnold Snyder’s “Blackjack
Forum”’ magazine in March 1995, reprinted with four pages
of new material in The Blackjack Shuffle Tracker’s Cookbook
in 2003.’ Snyder’s BJ Forum article was the first time anyone
had fully explained the technique in print.
Two other good sources are David Morse’s Blackjack
Reality’ (1997), which advises betting into “clumps” of Aces
and Tens, and John May’s Get the Edge at Blackjack’ (2000)
which includes a wide range of “sequencing” techniques.
Until now, a detailed book about Ace Prediction had not
been written, despite the fact several people were capable of
doing so (the retired game protection expert Steve Forte is one
of them).
The Ace Prediction “cat” was half way out of the bag so
I decided to let it go, accompanied by a gentle reminder that,
unless you are skillful, Ace Prediction could be hazardous to
your bankroll! However, there is a lot of life left in this
amazing technique. Study this book, then go out and check
the games for yourself using my analysis methods. You
will find gold in the most unexpected places.
22 Ace Prediction
PUBLISHED REFERENCES
The Gambling Times Guide to Blackjack (1983)
Dr. Edward O. Thorp mentioned Ace Prediction in Chapter

3 of this book: “One [technique] called ‘localization,’ keeps
track of ten-ace-rich regions of the used cards, if they are not
dispersed in the shuffle ... Another, which I published in a
paper on non-random shuffling, explains how to predict the
appearance of aces shortly before they are dealt (one-deck
game only).””
Man vs. Computer: Does Casino Blackjack Differ From

Computer-Simulated Blackjack? (1988)
Arnold Snyder, blackjack author and publisher, with Dr.

John M. Gwynn Jr.,'' (1932-2001), Professor of Computer
Science at California State University, Sacramento, wrote:
“Shuffle tracking’ and ‘card location’ are advanced strategies,
based on card-counting theory, and are most often employed
in conjunction with card counting. These tracking and
location strategies do take advantage of ‘non-random’ human
shuffles.”
Ruffled by the Shuffle (1990)
Snyder added in this Blackjack Forum article: “By limiting

their shuffles to a few riffles, players do preserve sequences of
cards, the knowledge of which can be exploited by
sophisticated players. I do know blackjack players who
exploit non-random shuffles in casinos. They use various
methods of shuffle tracking and card location strategies (based
on pre-shuffle sequences).”"
Card Counting for the Casino Executive (1990)
Bill Zender, who part-owned the Aladdin Casino in the

1990s, wrote: “Locating the value of cards in a shoe is not
Ace Prediction 23
that difficult. Cards can be glimpsed during the cut or during
the shuffle. The hard part is being able to ‘steer’ that card into
our hand. The professional card player uses several techniques
to locate and steer cards to his advantage.”
How to Develop a Perfect Memory (1993)
Dominic O’Brien, a British memory expert, includes a

short chapter about blackjack in this book. Although “Ace
Tracking count tracking and sequence tracking” are
mentioned, he gives no detailed description of the techniques:
“Sometimes an inexperienced croupier won’t shuffle a shoe
thoroughly. Imagine the advantage you would suddenly have
fe ols le memorized sequences of cards from the previous
shoe.”
One Step Ahead (1995)
Michael Konik, gambling columnist for Cigar Aficionado

magazine, wrote: “By memorizing three card sequences—
usually two ‘key cards’ preceding or surrounding the ‘target
card’ (an ace)—the accomplished sequence tracker can collect
enough data to follow the progress of desired cards through
the casino’s not-at-all-random mixing procedure.”'®
Card pros count more than money (1995)
Roseann Danese wrote in The Windsor Star, Ontario:

“Key girls’ are responsible for tracking aces and then secretly
signaling to their partner, the ‘big player’ ... If the system was
working—which would have depended upon several factors,
the most important being how they were shuffled by the
dealer—they would have been able to predict when the ace
emerged based on the cards that came out before it. Then, the
signal would be given that the ace was on its way and that the
bet should be pushed up to the maximum that the table
allows.’””!’
24 Ace Prediction
Unlocking the Shuffle: Trade Secrets of the Shuffle Trackers
Snyder offers eleven pages on “Sequential Tracking” in

part HI of his seminal Blackjack Forum Shuffle Tracking
Series. He writes: “These strategies have little value for solo
players. These are primarily team strategies, and for the most
part high stakes team strategies, as you would optimally prefer
to take over the table. One or two civilians at the table—who
are not part of your team—can neutralize your advantage too
easily.”®
Best Blackjack (1996)
In a section entitled “Tracking Aces,” Frank Scoblete

wrote: “One of the premier advanced techniques is to follow
aces through the shuffle ... Anytime four, five or more aces
come out in a given round—quickly memorize all the cards
that have come out with them; or, if that is too daunting a task,
memorize the several cards nearest the aces ... These are
called ‘key’ cards, as they will key you into the locale of the
aces if they start to appear after the shuffle.”
Fleecing Las Vegas (1997)
Michael Angeli’s racy 6,137-word Esquire article tells of a

blackjack team that beat Caesar’s Palace, Las Vegas, for
more than half a million dollars: “Players track the aces, as
many as they can. (Aces, remember, are important: You can
get blackjack only with an ace, and blackjack pays three to
two.) They do this by memorizing the three cards before an
ace: for example, king of hearts, 2 of clubs, 10 of spades, ace.
Somewhere in the game, that king, 2, and 10 are going to
appear, and they’ll know what's coming next: the ace. If they
can sequence three to five aces in a six-deck shoe (counters
trained in memory retention can sequence twice that), they’Il
have themselves a nice little betting advantage.”””
Ace Prediction 25
How to Make $100,000 a Year Gambling
for a Living (1997)
Professional gamblers David Sklansky and Mason

Malmuth wrote: “This technique will help you predict when
Aces will appear ... Knowing this can give you a significant
edge over the house and an expert Ace tracker, under ideal
conditions, can actually do better than an expert card counter.
However, notice that we said under ideal conditions.”
Blackjack Reality (1997)
In this self-published book David Morse gives eight pages

of detailed instructions for using his “signature card”™
method to predict strings of Aces and Tens. He writes: “[The
method] helps the player identify favorable clumps and to take
advantage of them, primarily by increasing the size of the bet
and secondarily by modifying the play of the hand.”
Baccarat for the Clueless (1998)
John May, a professional gambler, wrote: “This method

[Sequential Shuffle-Tracking] is very powerful indeed. It can
be used in any card game where the shuffle is nonrandom ... If
this method were widely publicized, I have no doubt the
casino industry would have serious trouble remaining
financially viable.”
Get the Edge at Blackjack (2000)
May provides a further ten pages on “card sequencing.” In

Chapter 6, Advantage Play, he explains: “These techniques
are the most lethal weapons in the arsenal of the advantage
player. Practiced correctly in the hands of disciplined
professionals they will decimate the bottom line of a casino in
a comparatively short space of time.“
26 Ace Prediction
Card Sequencing (2001)
Gambling expert Henry Tamburin wrote: “This technique

involves memorizing a key card that goes into the discard tray
before an ace, and then watching for the appearance of the key
card after the shuffle knowing that when it appears the ace will
not be far behind ... Once a card sequencer has memorized the
key card and it appears on the table, he will spread to two or
three hands with big bets to ensure that he is dealt the ace.””°
Aces & Faces Blackjack (2001)
Rob McGarvey’s short chapter “Card Sequencing: Smart-

Bomb Accuracy” states: “With Card Sequencing, we can tell
the exact location of cards and the order they will come out of
the shoe ... This is not a required skill to play Blackjack, but it
is a definite opportunity to make a killing at the tables ... Card
Sequencing works with simple riffle and re-stack shuffling,
and oe are lots of games out there being shuffled this
way.”
Clump Reading and Shuffle Tracking (2001)
An article by “Ace” on the ace-ten.com website states:

“The ability of a player to increase his wager when he knows,
as an absolute certainty, that an ace will be dealt to him in his
next hand, will definitely shift the odds in his favor. Without
knowing the value of any other card, there is a 31% chance he
will draw blackjack (and receive a 3:2 payoff)—and even if he
does not, there is a high probability he will be able to draw to
17 or better without busting.”
How to Win at blackjack (2002)
This short article, on the Observer Sport Monthly website,

reads: “Ace tracking” relies on the fact that when a hand is
over, the croupier scoops up the cards in a smooth, predictable
all
motion, in which each card slides under its neighbor and
94)
Ace Prediction
are put into the discard box in the exact order in which they
were originally dealt. An ace tracker memorizes the two cards
that go underneath each ace.”**
Reader of the pack (2002)
Sandra Newman’s humorous OSM website article tells of a

blackjack team that loses $4,000 at Atlantic City’s TropWorld
Casino, despite winning $1,500 on a single round by
predicting an Ace that receives a Ten as second card. They
move to Mississippi only to get barred. Finally, they travel to
Casino de Genting, in Malaysia. In several months of play,
they net “more than £200,000”? tracking Aces.
Bringing Down the House (2002)
‘Ben Mezrich tells the story of six Massachusetts Institute

of Technology students who use “keycard sequencing,”
shuffle tracking and card counting to beat the casinos for more
than three million dollars. In an essay on technique, team
member Kevin Lewis (not his real name) writes: “ ... we
would track the exact location of aces within the shoe and
memorize sequences of cards surrounding those aces. Then
we could predict when whole sequences would come out of
the deck.”
Hacking Las Vegas (2002)
In this Wired website article about the MIT blackjack team

Mezrich adds: “Along with tracking groups of high or low
cards, a trained counter can spot individual aces or even series
of aces. Since drawing an ace adds roughly a 37 percent
advantage to the player’s expected take, tracking a series of
aces through the shuffle can be extremely profitable.”*'
Card Shuffling Shenanigans (2002)
Ivars Peterson wrote in Science News: “In a game such as
28 Ace Prediction
blackjack, an astute player can try to memorize the cards
already played to have a better chance of predicting which
cards will come up later, thus potentially gaining an advantage
over the dealer and the casino. If the cards aren’t properly
shuffled and their distribution isn’t truly random ... certain
sets of cards may remain close enough together to be tracked
through the deck, and players can use such vestiges of pattern
to their profit.”
The Blackjack Shuffle Tracker’s Cookbook (2003)
A reprint of Snyder’s complete Blackjack Forum Shuffle

Tracking Series with four pages of previously unpublished
information on “sequential ace-location strategies” including:
False Keys; The Cost of Steering; The Value of an Ace; and
Lay-and-Pay vs. Pick-and-Pay (dealer procedures for picking
up the cards at the end of a round).*°
He had the upper hand, now all bets are off (2003)
Eliot Jacobson, a professor of computer science at the

University of California, Santa Barbara, tells how he was
barred from Chumash Casino Resort, Santa Ynez, after telling
the pit bosses he could predict the order of the cards: “I could
predict the aces with great accuracy. I once doubled-down on
a hard 20 (two face cards) to prove my point ... I got the
ace.”
Ace location techniques (2004)
Jacobson wrote on www.cardcounter.com.: “Knowing you

will be dealt an ace as your first card gives you over a 50%
edge, so trying to steer a known ace can be a very powerful
tool. However, the opportunities to use this kind of skill are
rare—they depend on an easy shuffle, having the round end
between the key card and the ace, and then controlling the
table enough so that you (or a team mate) gets the ace. Learn
how to do it for a few easy shuffles (stepladder, 1 or 2 pass
Ace Prediction 29
R&R), and if you get the chance, pounce!””°
“HOw I INVENTED ACE PREDICTION”—ED THORP
6 6 Ts big thrill,” said Ed Thorp, “came from learning

things nobody else in the world had ever
known.”’° Edward Oakley Thorp was a 28-year-
old Assistant Professor at New Mexico State
University when he came up with the idea of Ace Prediction:
“IT believe that I began to think in detail about non-random
shuffling in 1961 and 1962. My initial thoughts were that it
could very substantially affect the odds of many games.
“This was confirmed by the
subsequent work I did. I had a two-
pronged attack: build mathematical
models to approximate real shuffling,
and do empirical studies of real
shuffling. While doing this, I wanted a
simple, practical method for exploiting
this and the idea of Ace locating, using neighboring cards,
occurred to me. Why Aces? Because an Ace is the best card
for the playerto get as one of his initial two cards at
Blackjack.
“I tried it out at home and it worked well. I didn’t focus on
using it at the casinos because many other projects with higher
priority were going on in my life at the same time.”
Among Thorp’s “other projects” were inventing with
Claude Shannon?’ (1916-2001) the world’s first wearable
computer to successfully predict roulette outcomes in Las
Vegas, and writing the world’s best-selling gambling book,
Beat the Dealer,** which contained the first mathematical
system ever discovered for beating a major casino game—card
counting at blackjack.
Paul O’Neil, writing of Thorp’s exploits in Life magazine
in 1964, observed: “Thorp delved into a stratum of impure
chaos—a phenomenon involving both pattern and lack of
30 Ace Prediction
pattern—to which comparatively little attention had ever been
paid.””?
While Thorp’s book made the New York Times best-seller
list, his Ace Prediction theory remained the closely guarded
secret of a handful of high-stakes professional blackjack
players for more than 20 years.
The idea of predicting Aces first appeared in Thorp’s 1973
academic paper, “Nonrandom Shuffling with Applications to
the Game of Faro. | in which he wrote: “ ... nonrandomness
ce
yields simple winning strategies at Blackjack, Baccarat and

Faro (Thorp and Walden, unpublished)" ... ” After outlining
his non-random shuffling theories, Thorp commented: “Note
... the immediate application to ace-location ...
In 1997, Arnold Snyder, blackjack author and publisher,
asked Professor Thorp if he had explored non-random
shuffling more thoroughly in the seventies. Thorp replied: “I
did. Most of it never got written up. It is mostly in my mind
and in a few sketch notes.”
Richard Reid, Webmaster of www.bjmath.com, (a website
to which Thorp contributed) asked the professor if his “simple
winning strategies” would ever be published. “They may,”
Thorp replied, “appear in a future reprint of [his book] The
Mathematics of Gambling.”
Ace Prediction 31
NON-RANDOM SHUFFLING
PERMUTATIONS BY CUTTING AND SHUFFLING
Dr. Solomon W. Golomb, Professor of

Electrical Engineering and Mathematics at the
University of Southern California, Los Angeles,
analyzed movements of cards through successive
De perfect shuffles of a 52-card deck in his cnt)
release, “Permutations by Cutting and Shuffling” (1961).”
Perfect shuffle; also called the “out” shuffle
A deck of cards is divided into two equal halves and the

cards are riffle shuffled, alternately interlacing one card from
each half. Table 2-1 shows the first (top) and 52nd (bottom)
cards remain unchanged in position.” One perfect riffle of a
deck in initial order 1 to 52 produces the final order (reading
across):
*
One card separates a card in the final order from its follower in the initial order.
Eight perfect riffles will restore the deck to its initial order 1 to 52.
32 Non-Random Shuffling
TABLE 2-1. A PERFECT SHUFFLE
1 Py Di Es 8s ET PK)
5 31 6 32 of, 33. 834
9 35 10 SO pall Sit IP BS
13 39 14 40 PS S4ie iG 42
17 43 18 44 19 45 20 46
21 47 22 48 23 49 24 50
25 5] 26 52
Modified perfect shuffle, also called the “in” shuffle
A deck of cards is divided into two equal halves and the

cards are riffle shuffled, alternately interlacing one card from
each half. Table 2-2 shows the first (top) card moves to
position 2. The 52nd (bottom) card moves to position 51.*
One modified perfect riffle produces the final order:
TABLE 2-2. A MODIFIED PERFECT SHUFFLE
229, 3: 30. 4
CSS ea 634 8
ROM ES 7a ell: TS8r Pelee
A eels 425 6
oS Oe Ome 0)
22 AD 23 O24
26
Dr. Peter A. Griffin*® (1937-1998) Professor of

Mathematics at California State University,
Sacramento, wrote in The Theory of Blackjack
(Revised and expanded 1981): “If the dealer
: A performs a perfect shuffle of half the deck against
the other half, then, of course, the resultant order is
3946
deterministic rather than random.
the initial order.

* One card still separates a card in the final order from its follower in
modified perfect riffles will restore the deck to its original order | to 52.
Fifty-two
Non-Random Shuffling 33
Imperfect shuffle, also called the “amateur” shuffle
A deck of cards is cut at random near the center, and the

cards are riffle shuffled, alternately interlacing packets
consisting of one, two, three or more cards. After one
imperfect riffle, the number of cards separating a card in the
final order from its follower in the initial order can vary from
zero up to four or more.
Pat Macaluso, in his book Learning Simulation Techniques
on a Microcomputer Playing Blackjack and Other Monte
Carlo Games, wrote: “We have all seen and undoubtedly
admired the skilled card dealer who cuts a deck, smoothly
riffles the two halves together and continues in this way to
shuffle the deck. Beware! What you are witnessing is not
shuffling. It is a controlled reordering of the deck. If done
flawlessly, it will restore the deck to its original order after a
few such cuts and riffles. Even when it is not carried off
perfectly, the deck will retain much of its original order.”*”
THORP’S NON-RANDOM SHUFFLING ANALYSIS
In his classic blackjack book Beat the Dealer:

A Winning Strategy for the Game of Twenty-One
(1962), Dr. Edward O. Thorp wrote: “It is well
known that the order of a shuffled deck may be
very noticeably dependent on its previous order if
the shuffling is done either very poorly or very cleverly.”
Thorp cites Dr. William Feller” (1906-1970), Professor of
Mathematics at Princeton University, New Jersey, who wrote:
“It is seen that continued shuffling may reasonably be
expected to produce perfect ‘randomness’ and to eliminate all
traces of the original order. It should be noted, however, that
the number of operations required for this purpose is
extremely large.”*”
After publishing Beat the Dealer, Thorp
began writing a new book with William E.
Walden, his Ph.D. student at New Mexico
State University, Las Cruces.
Thorp was Assistant Professor of
Mathematics and chairman of the committee
that examined and approved 34-year-old
Walden’s 113-page doctoral dissertation,
“Solution of Games by Computation.” The
new manuscript, “The Solution of Games by
Computer,’ was being written with the
assistance of Julian H. Braun*” (1929-2000), a computer
programmer at the IBM Corporation. It contained Thorp’s
winning blackjack card counting systems and Walden’s
strategies for Nevada Baccarat and the oriental board game,
Go. Ina section entitled “Card Shuffling” Thorp discusses the
theory of computerized and human = shuffling, but
“Experimental” and “Applications” aspects of human
shuffling and “Questions for Further Investigation” were never
completed. The manuscript remains unfinished and
unpublished.
Thorp did go on to explore non-random shuffling more
thoroughly, beginning with the academic paper “Nonrandom
Shuffling with Applications to the Game of Faro,” (1973) in
which he wrote: ‘Since human shuffling seems intuitively to
be, and in fact is, decidedly nonrandom, it is surprising that
previous work on nonrandom shuffling seems to consist only
of Thorp and Walden (unpublished), of Epstein (1967) which
includes Epstein (1964), and Knuth (1969).°° It is even more
surprising because the play of many games Is significantly
altered when one considers the nonrandomness of human
shuffling ... The general analysis of nonrandom shuffling, and
its application in particular instances, is complex and
extensive. We hope to present it subsequently ... a
In exploring non-randomness in card shuffling, Thorp
suggested a simple mathematical model of a riffle shuffle.
Consider a four-card deck, all Spades, in the initial order Ace,
Two, Three, Four. The cards are identified as c/, c2, c3, and
c4 according to their location in the initial order. The deck is
cut into two equal halves, with c/ and c2 in one half and c3
and c4 in the other, and the parts are riffled together once. Let
N= 2. Leti=1. Cards c; and cy+;, at locations i and N+/
respectively, vie for locations 2i - | and 2i. In words, the Ace
of Spades (c/) and the Three of Spades (c3), at pre-shuffle
locations | and 3, vie for locations 1 and 2 in the final order.
The probability” of either card falling first, thus moving to
location 27 - 1 = 1 in the final order, is 0.5. Similarly, the
probability of either card falling second, thus moving to
location 27 = 2 in the final order, is 0.5. The cut and riffle may
produce any of 2N = 4 permutations,’ each with probability 2
"= 0.25. Table 2-3 shows the four possible permutations:
TABLE 2-3. CARD PERMUTATIONS
1,3, 2,4 1,3, 4,2 3,1,2,4 Bp

(ae:es
In a perfect riffle shuffle, where each card is interlaced one

card at a time, it is impossible for the Ace of Spades (c/) or
the Three of Spades (c3) to move to post-shuffle locations 3 or
4. The matrix P below shows the transition probabilities (P; ;)
that a single shuffle operation s will send a card in a specific
pre-shuffle location 7 to a specific post-shuffle location /.
TABLE 2-4. CARD TRANSITION PROBABILITIES
Pa* ipe 1 2 3 4
1] Os 90s 0 0
ce? 0 0 0.5 0.5
3 05 0.5 0 0
4 0 0 C5ant 05
It can be seen that, given the pre-shuffle location i of a

particular card, its approximate post-shuffle location + ARS
readily predictable. For example: Let i= 1. The pre-shuffle
location 7 of the Ace of Spades (c/) = 1. After a single riffle
* [FS] p. 188.
T FS] p. 186.
of the deck, the Ace of Spades can occupy one of only two
possible post-shuffle locations. It is either at location 2i - 1 =
1 or location 27 = 2. It cannot be at post-shuffle locations 3 or
4. In order for the deck to be randomized, all probabilities in
the matrix must be positive and equal, which means all cards
have the same chance of moving to any location in the final
order. For this simple four-card deck this happens after only
two riffles, but for larger decks where, for example, N = 26 or
more, several riffles are required to truly randomize the deck.
Thorp wrote: “In particular, when 2N = 52, P still contains
numerous zeros after five shuffles ... Although six shuffles
eliminate the zeros from P, six shuffles do not necessarily
produce all permutations with positive probability.”
EPSTEIN’S ANALYSIS OF THE IMPERFECT SHUFFLE
Richard A. Epstein investigated the nature of shuffles

performed by Las Vegas dealers and found a distribution with
“single-card interlacings approximately eight times as frequent
as two-card interlacings; a group of three cards appears less
than once per shuffle.”°° He represented the human shuffle
operation, s,, using the geometric distribution:*
Ey eed an
Where r = 1,2,3...
The probability of occurrence of a packet of length one

card, designated p, was approximately 8/9, and the probability
of nonoccurrence of a packet of length one, designated q, = | -
p, or approximately 1/9. The variable length of card packets
was denoted by 7. Substituting 7 for r in Eq. 2-1 yields the
probability P(7) that a sequence selected at random consists of
7 cards:
P(n) = (8/9)(1/9)""
Where 7=1);2,3:..
* TFS] pp. 166-167.
Non-Random Shuffling a7
Epstein pointed out: “It is evident from this equation that a
large measure of orderliness is preserved for a small number
of shuffles. This fact suggests the feasibility of predicting the
position of a card following the s, operation.” ’ For an infinite
deck, 7 can take any number of values (77 = I, Doan bute ter
a finite-length deck, 7 can take the values 1, 2, 3 only:
n=1,2 iia, n=4 2-3

P (n) =(8/9)(1/9)™" P(n) =(a/9)™" P(n) =0
Substituting the allowable values of 77:
P(1)= (8/9)(1/9)° P(2)= (8/9)(1/9)' P@B)= (1/9)? Daa

= (8/9)(1) = (8/9)(1/9) = (1/9)(1/9)
= 8/9 = 8/81 = 1/81
These fractions, designated r;, r. and 73, respectively,

represent the probabilities that a packet selected at random
from the final order will consist of one, two, or three
neighboring cards from the initial order. At this point we
make a distinction between two cards being physically
interlaced between two other cards, as with the riffle-shuffle
pattern A|BB]A, and two cards being “separated” by what
might be called “zero” cards, as with the pattern AA|BB. This
is important in Epstein’s analysis because he calculates not
only the probability of occurrence of different packet lengths
but also the probability of transition between packets, 1.e., the
likelihood of one packet’s ending and another’s beginning. In
the first case, two transitions occur between the packets, but in
the second, only one. Such transition probabilities are central
to Epstein’s development of a pragmatic card prediction
theorem.
When calculating 7 above, zero-card interlacings were
grouped with one-card interlacings. Since, in practice, they
are identical to two-card packets, we deduct 8/81 from r; and
add it tor. The probability of one-card packets becomes: 8/9
- 8/81 = 64/81 = 0.78; two-card packets = 8/81 + 8/81 = 0.20;
three-card packets = 1/81 = 0.02; (r; +1 +73 = 1). A second
empirical study of 40 riffle shuffles by four professional Las
Vegas dealers, performed in 1987 by Anthony Curtis, the
editor of Las Vegas Advisor, produced a lower probability of
one-card packets (0.66).°°
CASINO CARD SHUFFLES: HOW RANDOM ARE THEY?
Dr. Robert C. Hannum, Associate Professor

of Statistics at the University of Denver,
Colorado, did a thorough scientific analysis of
over one hundred casino card shuffles executed
by two professional Las Vegas dealers. He
analyzed 6 different six-deck shuffling
procedures at selected major Las Vegas Strip casinos in early
Loe
He concluded: “ ... six-deck shuffle procedures generally
are not random, failing virtually all tests for mathematical
randomness. For some procedures this nonrandomness is both
material and significant; for other procedures it is not clear
that this departure from randomness can be exploited or how it
might affect the play of a game.”°” Hannum’s probabilities for
various card packet lengths in a one-riffle shuffle are
compared with Epstein and Curtis’ results in Table 2-5:
TABLE 2-5. CARD PACKETS OF VARIOUS LENGTHS
1.00 ste
EES Epstein Hannum Curtis wee eae

Length 2 Bie ] Curtis
n Prob. Prob. Prob. 3 a0 |
a 0.304
i} 0.79 0.77 0.66 oeaul
0.00 ~—_——_— :
2. 0.20 0.18 0.26 1 2 3 4 5
Card Packet Length
3) 0.01 0.04 0.05
‘i ee om oa? FIGURE 2-1. CARD PACKETS OF

: une O08 oe VARIOUS LENGTHS
Totals 1.00 1.00 1.00
Hannum’s results for one, two and three riffles are compared
in Table 2-6:
TABLE 2-6. HANNUM’S SHUFFLE RESULTS
foe 1-riffle 2-riffle 3-riffle

n Prob. Prob. Prob.
It ONT 0.92 O10 Probability
2 0.18 0.07 0.03 2909999900>
NH04hDNANWOO
SO]
lolelelelolelolololelo}
3 0.04 0.01 0.00 = nN 3 4
4 0.01 0.00 0.00 Card Packet Length

Totals 1.00 1.00 1.00
FIGURE 2-2. HANNUM’S SHUFFLE

RESULTS
David C. Emanuel, of Oak Brook, Illinois, and Dr.

Kenneth H. Sutrick, Associate Professor of Computer
Information Systems at Murray State University, Kentucky,
suggested some practical ways to exploit non-random shuffles
in their paper “Non-Random Shuffling Strategies in
Blackjack” (1988).
They wrote: “ . casinos do not allow their dealers to
spend an inordinate amount of time shuffling cards. In many
casinos, shuffling may appear to be thorough, but merely
involves repeated mixing of the same two quadrants of the
shoe together. Astute blackjack players have noted this
tendency and exploited the information contained in a prior
round of play.”
INFORMATION LOSS IN CARD SHUFFLING
An interesting question is: how quickly is

information about the order of a deck destroyed
5e by successive riffle shuffles? Dr. Edgar N.
Gilbert, of Bell Laboratories, in Murray Hill,
New Jersey, proposed a mathematical model of
riffle shuffling that closely approximates
human shuffling.” The deck is cut roughly in half and the two
packets are riffled together. Given that all cards must
originate in one packet or the other there are 2°” possible final
orderings. The probability that the deck is in any specific
post-shuffle order 7 is denoted by p;.
To begin with, let us assume we know the exact order of a
deck of 52 cards prior to it being shuffled (p; = 1 for a single
i). Shannon’s formula for informational entropy® gives the
amount of uncertainty U associated with this situation:
U= 2 Pi log> Pi yes)
i=]
ieee eee log (eI)
As we might expect, since we know the exact order of the

deck, there is no uncertainty at all. Conversely, the amount of
information (in binary digits, or bits) is given by:
I= log,(n!) - U 2-6
Where,
n=52
U=0
I= log,(52!) - 0 = 225.58 bits
Starting with a deck in known order, each successive riffle

reduces the percentage of known information to the levels
shown in Table 2-7.
TABLE 2-7. INFORMATION LOSS IN CARD SHUFFLING
Riffles Info%
=
0 100
76.95
53.90
30.98 (%) SNWLhUOONWOOD
Information
lelelelelalelealelele}
12.09 Oe ee ee ee
B52 Riffles
0.92
0.23 FIGURE 2-3. INFORMATION LOSS IN CARD
0.06 SHUFFLING
G0)
NSF
(Oy
SS)
Aor
a7 0.01 os ;
(eyS
0.00 After the first riffle, log.2° = 52 bits of
information about deck order are destroyed
(23.05%) and 173.58 bits remain (76.95%). Reductions of
similar magnitude occur for the second and third riffles. After
the fourth riffle, only 12% of the original information
remains. As information is lost, uncertainty increases. After
ten riffles, J= 0 bits and U = log,(52!) = 225.58 (p;= 1/n! for
all i).
In explaining his card prediction theorem, Epstein
remarked: “To apply this observation usefully, we consider
only those transition processes comprising one, two, or three
shuffles; for ... six or more shuffles, the statement is virtually
useless for a conventional deck.”
This suggests players attempting to predict the card order
in shuffled decks should limit themselves to blackjack games
with a maximum of three riffles. With a simple three-riffle
shuffle, enough information remains to make the post-shuffle
deck order reasonably predictable. Four or more riffles are
too unpredictable.
Measuring the degree of non-randomness in a shuffle is
more difficult. We might, however, usefully compare
imperfect riffles with perfect riffles. For example, in a perfect
riffle the deck is broken at exactly 26 cards. A one-card
interlacing occurs every time (probability = 1) and a two-or-
more-card interlacing never occurs (probability = 0).
Using this as our yardstick, we can judge how close a
dealer’s riffle is to perfect. Some authors have referred to a
dealer's unique riffling pattern as that dealer’s signature.®°
The empirical shuffling studies presented in this chapter show
that professional dealers drop single cards 70 to 80% of the
time, and rarely drop more than two cards.
A MAN OF MYSTERY: CHARLES T. JORDAN
ce Prediction is based on a shuffling secret discovered

early last century by a young American card trick
developer from Penngrove, California, named Charles
Thorton Jordan®’ (1888-1944).
Jordan, who was also a “designer of radios, professional
contest winner and chicken farmer,’ would type out the
instructions for performing his card tricks, then print,
assemble and bind the pages into pamphlets with his own
hands, before selling them using mail order advertising in
conjuring magazines like Floyd G. Thayer’s Magical Bulletin
(1914-1925).
In 1919 Jordan privately printed his classic booklet Thirty
Card Mysteries and, in Chapter 1, Trailing the Dovetail
Shuffle to Its Lair, he typed: “My first intimation that a
dovetail shuffle was not a shuffle in the strictest sense of the
word, occurred a good many years ago. After shuffling a new
pack once in that manner, I chanced to glance through the
face-up pack; and what I saw caused my eyes to open. I cut
the deck, and looked again: then I pondered- - ae
Jordan went on to construct many baffling card effects to
exploit the phenomenon he noticed: that, even if not
performed perfectly, dovetail or riffle shuffles do not disturb
the order of the cards.
Another respected inventor of card magic, C. O.
Williams” (1880-1937), of Cardiff, Wales (who performed as
“Charles Oswald”) allowed his magician friend Angelo John
Lewis (1839-1919), known as “Professor Louis Hoffman,” to
publish a card trick in 1912 that used the same basic premise,
but Jordan’s writings “are considered the outstanding
treatment of the subject.””!
Williams prearranged a deck with “one-way” card-back
designs so they were all aligned in the same direction, and
then riffle-shuffled the cards once. As he shuffled, he turned
half of the deck in the opposite direction, so that two rising
sequences could be clearly identified and a magical “card
reading” performed.”
Mystery still surrounds the reclusive Jordan. Thomas
Baxter, a Canadian illusionist, and “Marko,” editor of The
Learned Pig Magic eZine, wrote of a visit to Penngrove by W.
F. “Rufus” Steele, a famous Chicago card magician: “He
found out Jordan’s place was three miles out of town and there
he went.
“What he found was an unpainted shack, a partly collapsed
barn, unwashed dishes all over, and chickens flying in and out
at their pleasure and strolling across the table. He also found a
‘farmer type’ who told him he was Jordan but he had never
invented a trick in his life.
“Instead he sold the effects for a man in New York who,
apparently, did not want his name associated with magic.
Farmer Jordan took out a large envelope with a batch of
typewritten instructions for the tricks. The name of the man
who had sent the envelope was Arthur Finley.”’”*
ANALYZING SHUFFLES
APPLYING THORP’S SHUFFLING ANALYSIS
Thorp (1973) provided a simple

procedure for analyzing the
distribution of cards in a shuffle.”
A new, standard 52-card deck of
Bee® casino cards (No. 92 Club
Special, Match Blue Diamond Back
No. 67) was used in _ this
experiment, which was performed
by the author.
Methodology: Numbers | to 52
were marked on the card backs.
The cards were arranged in order
from 1 to 52, giving the initial
order. The deck was cut and riffle-
shuffled three times. After each riffle the numbers on the
cards were typed into a computer, giving a permanent record
of the final order. These steps were repeated 30 times for a
one- and two-riffle shuffle and 100 times for a three-riffle
shuffle. A tally was kept of the number of cards, in the final
* [FS] p. 161.
Analyzing Shuffles 45
order, separating a card c; from its follower c;+1 in the initial
order, which Thorp called D;.
Using this tallying method, where the number of cards
separating two other cards is being counted, it is technically
“possible” for no cards to be interlaced, in which case D; = 0.
In total, 51 X 30 = 1,530 observations of D; (n = 1,530) were
recorded for a one and two-riffle shuffle, and 51 X 100 =
5,100 observations (n = 5,100) for a three-riffle shuffle.
Therefore, the statistics presented can be looked upon as good
estimates of the parameters” of the parent group for all card
shuffles of a single deck where the shuffle consists of a simple
cut followed by a riffle, repeated once, twice or three times.
The raw distribution of D; includes broken sequences. A
sequence is broken when two cards that were neighbors in the
initial order are separated by a cut. In the shuffle analysis that
follows, the probability of broken sequences from riffle
shuffling and other shuffle actions like plugs, strips and box
cuts, is calculated collectively. This leads to double counting
of the broken sequences caused by the riffle shuffling.
Therefore, to use the data in future calculations we should
rightly adjust the raw distribution to exclude broken
sequences.
In the adjusted distribution, broken sequences are reduced
to zero. The adjusted probabilities are calculated by dividing
the raw values for each D; by the adjusted sample size. When
completed for all D; the effect is to scale the adjusted
distribution back up to a full 100%. This distribution has an
almost symmetrical “bell-shaped” normal curve.t This makes
the arithmetic meant and standard deviation’ appropriate
measures of location and dispersion, and allows the use of
normal probability tables for calculating confidence intervals**
for the unknown true population parameters.
* [FS] pp. 185-186.

1 [FS] pp. 183-184.
t FS] p. 178.
8 [FS] p. 198.
*“* [FS] pp. 150-151.
46 Analyzing Shuffles
RESULTS
Values of D; for A One-Riffle Shuffle
Fig. 3-1 shows that one card in the final order most
frequently separates two cards that were originally neighbors
in the initial order. The exact number of single-card
Separations may vary from person to person but, over a large
number of trials, the most frequently occurring distance
between cards after one riffle always will be one card.
TABLE 3-1. CARD GAPS FOR A ONE-RIFFLE SHUFFLE
1.00
D; n Raw Adj. 0.90 | | | |
0. 27 OAT hn 0181 a) |
= 0.60 | |
J 1059 0.692 0.686 eee,
2 176 0.115 0.117 oe
3 21 0.014 0.014 er ni ened ae a sesh
4 3 0.002 ~—-0.002 0 t 2
D,
3 4
Totals 1530 1.000 1.000

FIGURE 3-1. CARD GAPS FOR A ONE-
RIFFLE SHUFFLE
Summing zero and single-card probabilities gives a

combined probability 0.177 + 0.692 = 0.869. The probability
of two-card separations is 0.115. Dividing 0.869 by 0.115
gives 7.6. Therefore, single- and zero-card separations are
approximately eight times as frequent as two-card separations.
Three-card separations occurred 21 times in 30 shuffles, or
less than once per shuffle, and four-card separations were very
infrequent, occurring only three times in 30 shuffles.
The number of broken sequences was 30. The total sample
size was 1,530. Thus, the percentage of sequences broken was
(30 / 1,530) X 100 = 1.96%. Before dividing the total for
single-card separations, which includes broken sequences,” the
raw value was reduced by the number of broken sequences.
* .
Based on a “perfect” riffle.
Thus, the raw value for one-card separations (1,059) was
reduced by 30 to 1,029, before dividing by the adjusted
sample size, n = (1,530 - 30) = 1,500, to give the adjusted
probability 0.686.
Sample mode” = | card
Sample arithmetic mean (x)

Ler A/n
= 1 + (-0.029) = 0.97 cards
Where,
X0 (the assumed mean) = | card
A=-44
n= 1500
Sample standard deviation (s)

s=V((B-(47/n)(n- 1))
= V ((558 - (1936 / 1500)) / (1499)) = 0.61 of acard
Where,
B=558
A’ = 1936
n= 1500
95% confidence interval for the unknown true population

mean of D; (u)
Sample standard deviation s = 0.61 of a card

Sample size n = 1500
Confidence interval c = as / \n where a = 1.96
c= 1.96 * 0.61 / ¥1500= 0.031
95% confidence interval = 0.97 - 0.031 to 0.97 + 0.031, that

is, from 0.939 cards to 1 card. In 95 out of 100 such samples,
* [FS] p. 178.
the confidence interval will contain the true population mean
LL.

proportion of D; (P)
Sample population proportion p = 1029 / 1500 = 0.686

Confidence interval c = 1.96V(p(1 - p)/n)
c = 1.96\(0.686(1 - 0.686) / 1500) = 0.024

95% confidence interval = 0.686 - 0.0243 to 0.686 + 0.024,
that is, from 0.662 to 0.710. In 95 out of 100 such samples,
the confidence interval will contain the true population
proportion P.
Values of D; for A Two-Riffle Shuffle
Three cards in the final order most frequently separate two

cards that were originally neighbors in the initial order (Fig.
3-2). The number of broken sequences with a two-riffle
shuffle was 90. The total sample size was 1,530. Thus, the
percentage of sequences broken was (90 / 1,530) X 100 =
5.88%. In the adjusted distribution, broken sequences are
reduced to zero. The total sample size shrinks from 1,530 to
1,440. The adjusted distribution is calculated by dividing the
raw values for each D; by the adjusted sample size. Before
dividing the total for four-card separations, which includes
broken sequences, the raw value was reduced by the number
of broken sequences. Thus, the raw value for four-card
separations (301) was reduced by 90 to 211, before dividing
by the adjusted sample size, n = (1,530 - 90) = 1,440, to give
the adjusted probability 0.147.
TABLE 3-2. CARD GAPS FOR A TWO-RIFFLE SHUFFLE
oS Raw Adj.
74 ~=0.048 0.051
184 0.120 0.128
253 0.165 0.176
5500359 0.382
301 0.197 0.147
\oNn 0.062 0.066
47 0.031 0.033
17 0011 0.012 FIGURE 3-2. CARD GAPS FORA
4 0.003 (0.003 TWo-RIFFLE SHUFFLE
ey
SS)
[ey
Wey
LS)
KS
SNS
aS
a 4 0.003 0.003
10 0 0.000 0.000
11 1 0.001 0.001
Totals 1530 1.000 1.000
Sample mode = 3 cards

x=XxXot+A/n
=3 + (-0.13) = 2.87 cards
Where,
Xo (the assumed mean) = 3 cards
A151
n= 1440

s=V((B-(47/n))/(n-1))
= ¥ ((3249 - (32761 / 1440)) / (1439)) = 1.50 cards
Where,
B= 3249
A =3276)
n= 1440

mean of D; (1)
Sample standard deviation s = 1.50 cards
Confidence interval c = as / Vn where a = 1.96
c= 1.96 * 1.50 / ¥1440 = 0.077

is, from 2.793 to 2.947 cards. In 95 out of 100 such samples,
Ll.

Sample population proportionp = 550/1440 = .382

Confidence interval c = 1.96V(p(1 - p)/n)
c = 1.96V(0.382(1 - 0.382) / 1440) = 0.025

is, from 0.357 to 0.407. In 95 out of 100 such samples, the
confidence interval will contain the true population proportion
ee
Values of D; for A Three-Riffle Shuffle
Seven cards in the final order most frequently separate two

cards that were originally neighbors in the initial order (Fig.
3-3). The number of broken sequences with a three-riffle
shuffle was 700. The total sample size was 5,100. Thus, the
percentage of sequences broken was (700 / 5,100) X 100 =
13.72%. In the adjusted distribution, broken sequences are
reduced to zero. The total sample size shrinks from 5,100 to
4,400.
TABLE 3-3. CARD GAPS FOR A THREE-RIFFLE SHUFFLE
D = Raw Adj.
0 45 0.009 0.010
1 110 0.022 0,025
Probability
2 173 0.034 0.039
Se
o9e
Soe
eS°°o>
0.082
O-=NWhNONOOD
OCOOOCOOO000oO
3 361 0.071 012 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
D;
4 278 0.055 0.063
2) 439 0.086 0.100
FIGURE 3-3. CARD GAPS FORA
6 1338 0.262 0.145
THREE-RIFFLE SHUFFLE
7 830 0.163 0.189
8 507 0.099 0.115
The adjusted distribution is
9 326 0.064 0.074
calculated by dividing the raw values
10 241 0.047 0.055
for each D; by the adjusted sample
11 176 0.035 0.040
12 110 0.022 0.025 size. Before dividing the total for
13 59 0.012 0.013 six-card separations, which includes
14 38 0.007 0.009 broken sequences, the raw value was
its) 31 0.006 0.007 reduced by the number of broken
16 13 0.003 0,003 sequences. Thus, the raw value for
WET: 5 0.001 0.001 six-card separations (1,338) was
18 g) 0.002 0.002 reduced by 700 to 638, before
29 11 0.000 0.001 dividing by the adjusted sample size,
Totals 5100 1.000 1.000 n = (5,100 - 700) = 4,400 to give the
adjusted probability 0.145.
Sample mode = 7 cards
X=xXot+A/n
= 7 + (-0.27) = 6.73 cards
Where,
*0 (the assumed mean) = 7 cards
A = -1208
n= 4400
s=V((B - (47/n))/(n-1))
= V((41624 - (1459264 / 4400)) / (4399)) = 3.06 cards
Where,
B=41624
A’ = 1459264
n= 4400

mean of D; (1)
sample standard deviation s = 3.06 cards

sample size n = 4400
confidence interval c = as / Vn where a = 1.96
c= 1.96 * 3.06 / Y4400 = 0.090
95% confidence interval = 6.73 - 0.090 to 6.73 + 0.090 that is,

from 6.64 cards to 6.82 cards. In 95 out of 100 such samples,
Ul.

sample population proportion p = 830 / 4400 = .189

sample size n = 4400
confidence interval c = 1.96V(p(1 - p) /n)
c= 1.96V(0.189(1 - 0.189) / 4400) = 0.011
95% confidence interval = 0.189 - 0.011 to 0.189 + 0.011 that

is, from 0.178 to 0.200. In 95 out of 100 such samples, the
confidence interval will contain the true population proportion
Pp
BROKEN SEQUENCES
A sequence is broken when two cards that were neighbors

in the initial order are separated by a cut. For example, take a
single deck and number the cards 1 to 52. Cut it near the
center and riffle once. Let’s say the cut is at exactly 26 cards.
This means card 27 will be somewhere near the top of the
deck in the final order, preceding card 26. The original
sequence 26, 27 is now broken. In addition, the riffle has
created two interleaved rising sequences, the first numbered 1
to 26 and the second, 27 to 52. A second cut followed by a
riffle will break not one, but two more of the original two-card
sequences.
The probability of a broken sequence is much higher at
some locations in the deck. For example, the sequence at
locations 26 and 27 is much more likely to be broken than the
one at locations | and 2. Since the precise location of a
sequence is not easily obtainable in actual play, the shuffle
analysis that follows assumes all sequences within a single
deck have equal probability of being broken, regardless of
their location in the deck. In the simple riffle-and-restack
shuffle analyzed, a six-deck stack is cut and riffed repeatedly
to rebuild a final stack. The greater the number of cuts in a
shuffle the higher the probability a sequence will be broken.
For example, a shuffle where the six-deck stack is broken in
two and then the two stacks married to each other with only
two riffles of each grab will leave many more sequences intact
than a shuffle where the cutoffs are first triple-plugged, then
six piles are made, and each marriage is riffled three times,
followed by a second pass, etc. Players must learn how to
count the breaks as well as the number of riffles.
When a grab containing a two-card sequence is riffled
together with another grab, there are three chances the
sequence will be broken. The first is when the grab is lifted
off the pile for the first riffle. After one riffle, sequence length
expands, on average, by one card, and there are two more
chances of the sequence being broken. In two riffles, grabs
containing two sequences have six chances of being broken.
Where plugging the cutoffs, strip shuffling or boxing the deck
could cause sequence breaks, they also must be counted.
COUNTING THE BREAKS
In this section we analyze a two-riffle riffle-and-restack

shuffle (also known as R&R or Zone Shuffle) with the
following definition: six-deck, 66% penetration, one-pass,
four-pile, criss-cross, all half-deck grabs, cutoffs topped. Ace
trackers attempt to locate specific cards, so the number of
riffles performed as half-deck grabs are shuffled together is
critical information. In this shuffle, two half-deck grabs are
riffled together twice, with a single break between the riffles.
Four 2-card sequences (c/, c2), (c3, c4), (c5, c6) and (c7,
c&) are located in half-decks D, G, E, H, respectively. When
riffling expands the sequences, the cards are shown separated
by Xs, for example: c/,.X, c2.
Four decks of discards are

stacked in the discard tray. Starting
from the bottom, 1.e., the order in
which they entered the discard tray,
cards are identified in half-decks as
Segment A, Segment B, Segment
C, etc. Two decks remain in the
cutoffs. These cards are identified
as half-deck segments I, J, K and L.
The cutoffs have been topped, 1.e.,
placed on top of the discards, to form the
six-deck pre-shuffle stack. There are no
sequences in F but the grab may break a eg
¢3.c4 in.G. «Chance ,olc3,c4 beme Fee =
broken is 1/51. c5, c6
e1, 62
The stack is broken into

two 3-deck piles. The top 2s
half is placed on the table Break2a Break 34
on the right side of the c7, c8
bottom
:
half.
;
The3 halves oe Pile 2 Pile 1
are identified as Pile 1 and
Pile 2. The piles are broken into four 1'4-deck piles. The
quarters are identified as Pile 1, Pile 2, Pile 3 and Pile 4.
There are no sequences in I. Chance of Break 2 causing a
broken sequence is 0. There are no sequences in C but the
grab may break c/, c2 in D. Chance of c/, c2 being broken is
L/S1s
Dealer grabs F and I. There are no

sequences in F but the grab may
break c5, c6 in E. Chance of a c5, c6
being broken is 1/51.
There are no sequences in

I but the grab may break c7,
Eo uingh eChancesok G/arce c3, c4
being broken is 1/51.
F and I riffled together once. F+I is | F+! | €Break6

broken for a second riffle. There are no Se
: Final
sequences in F+I. Chance of a broken
sequence is 0. The twice-riffled deck F+I is placed on the table
to form the final stack.
Dealer grabs L and C. There are

no sequences an L “Chance of a
broken sequence is 0.
There are no sequences in

C/,
C. Chance of a broken 8
sequence 1s 0. es,
Pile 2 Pile 1
C and L riffled together once. C+L is & Break9

broken for a second riffle. There are no ar zeae
sequences in C+L. Chance of a broken
sequence is 0. Twice-riffled deck C+L Final
added to final stack.
Dealer grabs E. Chance of c5, i¢ Break 10

c6, or cl, c2 being broken is
fey
Dealer grabs H. Chance of Ha Wien ae Break 1 |(ee

4) |367 aan
e7, cé, or c3, C4 being broken
4
4, aij
isr2/ 50s
E and H riffled together once. ae E+H | & Break 12

E+H is broken for a second riffle. %e6 y 88
There are two sequences in E+H. 2” riffle
Chance of c5, c6 or c7, c& being eal
broken is 4/51. Twice-riffled
deck E+H added to final stack. Final
Dealer grabs B and K. a Break 13 4
There are no sequences in B. A G A
Chance of a broken sequence Pile 2 Pile 1
is 0.
There are no sequences in K. 4 ea . .

Chance of a broken sequence is 0. ip =dt
B and K riffled together pd ney

once. B+tK is broken for a EsH
second riffle. There are no Cie ae Pya oe Ds
sequences in B+K. Chance a ee *
of a broken sequence is 0. lees
Twice-riffled deck Bt+K
added to final stack. Final
Dealer grabs D. Chance of c/, c2 7 J ar

being broken is 0. Fick a Mac
Dealer grabs G. Chance of 304 [_G As
Pile 2 Pile 1
c3, c4 being broken is 0.
D and G riffled together

once. D+G is broken for a
second riffle. There are two
sequences in D+G. Chance
of el) e2 or <3, cf being
broken is 4/51. Twice-riffled
deck D+G added to final
stack.
Dealer grabs A and J. There are no ja
sequences in J. Chance of a broken T''e4
sequence is 0.
There are no sequences in A. lash

Chance ofa broken sequence is 0. an
A and J riffled together i Break17

once. A+J is broken for a as
second rifle. “There are no ct, % xx, C3 1G
sequences in A+J. Chanceof “ a
a broken sequence is 0. The
double-riffled deck A+J is Pa i
added to final. A player cuts. Wg kcs pe voniee
Player Cut. Assuming the Final

casino does not allow a player to cut less than one deck from
the bottom of the six decks, the chance of any sequence being
broken by a random player’s cut is 16 / (312 - 53) = 0.06.
PROBABILITY OF BROKEN SEQUENCES
Summing the fractions for each sequence:
TABLE 3-4. PROBABILITY OF A BROKEN SEQUENCE
Sequence OR
GING? eae 4/511
c3,c4— 4/51
c5,c6 = 4/51
c7,c& 4/51
Total 16/51
Analyzing Shuffles Ss)

The average probability p, of one sequence being broken
is 4/51 = 0.08. The probability p» of two sequences being
broken is 0.08 X 0.08 = 0.01, (p3 and p, = 0). Finally, 1 - (p,
+ p.+p3+ ps gives the probability po for zero sequences:
TABLE 3-5. PROBABILITY OF BROKEN SEQUENCES
Sequences AND 0.90 } a =

0.80 | = Pae
0 0.91 070 |
060
= Aer =
1 0.08 A >
2 01
0.0 2 0.40
ae eae ——
res
0.00 0.20 =
3}
0.10 ——
4 0.00 0.00 oa
Total 1.00 : : : ;
>] 0.09 Broken Sequences
FIGURE 3-4. PROBABILITY OF BROKEN

SEQUENCES
The average probability of one or more broken sequences

is 0.09. Adding the chance of a random player’s cut breaking
a sequence (0.6), the total probability of a broken sequence in
this shuffle is 0.15.
BROKEN SEQUENCES BY COMPUTER
The software Casino Veérité (CV) Shuffle provides a

percentage for intact sequences that can be used to calculate
the probability of broken sequences in any shuffle.
CVShuffle was set to count as “intact” only those
sequences where the first card and second card were separated
by no more than three cards. Thus, two cards separated by
four or more cards were not counted. The probability of
intact sequences for a simple, two-riffle shuffle of a single
deck was obtained from Table 3-2. Summing the adjusted
figures for values of D; in the range 0-3 card separations, the
probability of intact sequences was found to be 0.737. Next,
the above R&R shuffle, designated BJAP 1, was simulated
four thousand times using CVShuffle. The result is shown in
Table 3-6:
TABLE 3-6. BROKEN SEQUENCES BY COMPUTER
Intact sequences Prob.

30 simple two-riffle hand
shuffles ee
4,000 BJAP I computer shuffles 0.700
Difference 0.037
Add broken sequences 0.059
BJAP I broken sequences 0.096
For the more complex B/JAP / shuffle, the probability of

intact sequences fell to 0.700. The difference between the
simple two-riffle shuffle and BJAP J (0.737 - 0.700 = 0.037)
is added to the broken sequences figure for all two-riffle
shuffles (0.059) to give 0.096. This is in close agreement with
the 0.9 estimate calculated using pencil and paper above.
Using CVShuffle, the formula:
B= G1) =e Dy 3-]
Where,
B = probability of broken sequences
i, = probability of intact sequences for the hand shuffles
i, = probability of intact sequences for the computer shuffles
b;, = probability of broken sequences for the hand shuffles
can be used to quickly calculate the probability of broken

sequences for any shuffle.
EFFECT OF PLUGS, STRIPS AND BOX CUTS
Assume that, in our simple riffle-and-restack shuffle, the

dealer double plugs the two decks of cutoffs by splitting them
into two one-deck segments and inserting them into the
discard stack exactly one deck down from the top and one
deck up from the bottom. Also, for each deck in the six-deck
shoe, the dealer performs a strip shuffle just prior to
performing the /irst riffle.
For analysis purposes, we shall assume the stripping action
is of the thick variety, where 12-15 cards at a time are
“stripped” off the top of a deck and dropped on to the table,
one on top of the other, to form a rearranged deck. After the
second riffle the dealer performs a boxing action (a lopsided
cut) on each deck.
We can calculate the precise number of additional breaks
these shuffle actions will introduce. A sequence is subjected
to the risk of being broken if, and only if, the dealer performs
an additional cut or cuts on the segment in which it is located.
For example, plugging the cutoffs one-deck up from the
bottom has no effect whatsoever on cl, c2 in D, but plugging
the cutoffs one-deck down from the top could break c3, c4 in
G. Also, if the cutoffs had been plugged at any random point
in the discards (as often happens in real games) there is a risk
of any sequence being broken by a plugging action.
The strip shuffle as described above would involve three
additional cuts on a one-deck segment (at roughly card
locations 13, 26 and 39). If a half-deck contains one
sequence, the stripping action will increase the risk of a break
by 3/51. If there are two sequences in a half-deck this risk is
doubled.
Lastly, boxing involves one additional cut for each deck.
So, for example, if a half-deck contains one sequence, the
boxing action will increase the risk of a break by 1/51. If
there are two sequences, then, again, the risk is doubled.
The diligent sequence tracker will eyeball a shuffle closely
to determine if the dealer consistently breaks segments
containing tracked Aces. If so, the easiest solution is: go find
a better shuffle. Don’t make the common mistake of trying to
turn a bad shuffle into a good one. The additional plugging,
stripping and boxing increase the chance of a broken sequence
as shown in Table 3-7:
TABLE 3-7. EFFECT OF PLUGS, STRIPS AND BOX CUTS
Sequence Plugging Stripping Boxing

Giaie2 0 3/51 1/51
c3, c4 1/51 S/o 1/51
GeO) 0 3/51 1/51
CL e8 0 3/51 1/51
Totals 0 12/31 4/51
Adding these fractions to the ones given previously yields

Table 3-8:
TABLE 3-8. A BROKEN SEQUENCE ADJUSTED
Sequence No PSB PSB OR

Gclme2 4/51 4/51 8/51
c3, c4 4/51 5/51 9/51
c5, c6 4/51 4/51 8/51
G7, 68 4/51 4/51 8/51
Totals N6/SL U7/SL) 233/51
The average probability p,; of one sequence being broken

is about 8/51 or 0.156. The probability p, of two sequences
being broken is 0.156 X 0.156 = 0.024, (p; and p, = 0).
Finally, 1 - (p; + p2 + p3 + p4) gives the probability p, for zero
sequences:
TABLE 3-9. BROKEN SEQUENCES ADJUSTED
Sequences AND: i coach reo ae a

0.80 pete, Se |e eee See
0 0.83 , 0.70 css = = ae
dE OS = 0 ane ik SS
© 0.50 + = sss —
oe ceo ssCacia MEE ae ‘
2 0.02 1
630 (= Ee ae ee ae eee bs
3 0.00 0.20 E ae hae
On On == ete = —_
4 0.00 0.00
Total 1.00 a 1 2 3 4
Broken Sequences
>/ 0.17
FIGURE 3-5. BROKEN SEQUENCES ADJUSTED
Thus, the average probability of one or more broken

sequences is 0.17. CVShuffle was used to simulate this
shuffle, designated BJAP 2, four thousand times, and the
probability of intact sequences was found to be 0.620. Using
the formula given earlier, B was calculated as:
B = (0.737 - 0.620) + 0.059 = 0.176
Adding the probability of a random player’s cut breaking a

sequence (0.06) the total probability of one or more broken
sequences has increased from 0.15 to 0.23. The additional
cuts may have wrecked the profitability of this game.
AN UNCUT DIAMOND
GG ey, Laurent, listen to this!”

Laurent Bujold was an expert at cutting Aces out
of a shuffled deck. He tore his eyes away from
the café’s splendid view of Villefranche-sur-Mer’s port de la
Santé while I read aloud from the newspaper:
A RARE DIAMOND worth

100,000 francs lif] being
offered as a prize at the
blackjack table during the
maiden voyage of the new
luxury cruise ship, Palais
de la Méditerranée, which
sails for Monte-Carlo from
Nice tonight.
We immediately caught the train to Nice, where we found

the Palais garlanded with thousands of pink carnations in
honor of her first voyage. After boarding, we made our way
to the casino, and discovered the tiny colorless jewel nestling
snugly on a bed of black velvet under glass.
On the blackjack table were the words: ACE OF DIAMONDS
PAYS 100 TO 1. Laurent went to work and, in less than ten
minutes, had cut an Ace of Diamonds to the top of the six
decks. He placed a 1,000-franc chip on first base. In a flash,
the Italian dealer whipped the top card out of the shoe and
turned it over—it was a Four of Spades!
Laurent gasped, but before he could say anything the pit
boss announced we were to be barred immediately.
“But why?” asked Laurent.
“You are controlling the game,” he said curtly.
“We lost!” protested Laurent, but the boss refused to listen.
We had no choice but to go to our cabin.
In due course, there was a knock at the door and, on
opening it, a steward handed us an envelope. It contained an
invitation to dine at the Captain’s Table.
Dressed in our white tuxedos, we proved to be the only
guests of Captain Santorini.
“Welcome, gentlemen,” he said, as we took our seats. “I
imagine you’re mystified! It is quite simple. Our dealer,
Silvano, is—like you, Signore Bujold—a _ cardsharper.
Indeed, the top card was the Ace of Diamonds, but, on my
orders, he dealt you the card beneath. I doubt you will go to
the police. After all, you came here to swindle us out of the
diamond.
“Besides,” he added with a wry smile, “I’m afraid we’ve
beaten you to it. The jewel in the case is a fake. The real one
was sold in Tangier three weeks ago.”
EXPLOITING SHUFFLES
APPLYING HANNUM’S SHUFFLING ANALYSIS
Hannum (2000) provides simple procedures for testing

randomness in casino card shuffles. Six new, standard 52-card
decks of Bee® cards (No. 92 Club Special, Match Blue
Diamond Back No. 67), mixed together, were used in this
experiment, which was performed by the author. As a
prelude, the author’s single-riffle shuffle “signature” was
compared to those of the two professional Las Vegas dealers
who performed shuffles in the Hannum experiments. The
result is shown in Table 4-1:
TABLE 4-1. THE AUTHOR’S SHUFFLE SIGNATURE
Packet
Author Hannum
Length
n Prob. Prob.
Probability
1-card 0.83 0.77
85885233883
es0909099990>
2-cards 0.14 0.18
3-cards 0.02 0.04 Card Packet Length
More 0.01 0.01 FIGURE 4-1. THE AUTHOR’S

Totals 1.00 1.00 SHUFFLE SIGNATURE
66 Exploiting Shuffles
The author interlaces slightly more one-card packets and
slightly less two and three card packets, but the differences are
not large enough to affect the outcome of the experiment that
follows.
Methodology: Numbers 1 to 312 were marked on the card
backs. The cards were arranged in order from | to 312, giving
the pre-shuffle order. The cards were shuffled using a
common casino shuffling procedure known as the stepladder
shuffle (also known as stutter, dilution, staggered riffle shuffle,
or power series shuffle). A second pass using a simple riffle-
and-restack shuffling procedure followed this. Thus the
whole procedure comprised a two-pass combo shuffle. After a
single iteration the numbers on the card backs were typed into
a computer, giving a permanent record of the post-shuffle
order. The two-pass procedure was repeated for four
iterations.
The Two-Pass Combo Shuffle Procedure (BJAP 3)
1. The dealer breaks the pre-shuffle stack into four piles.
2. Half-deck grabs are made from two piles and riffled

together once to form a final stack.
3. A third grab is made from one (alternating) pile and a

fourth grab from the already shuffled final stack and
riffled together once before being placed back on the
final stack.
4. This is repeated until the first pass is complete.
5. The dealer breaks the final stack into two three-deck

piles. A half-deck grab is made from each pile and
riffled together once.
6. This is repeated until the second pass is complete.

The cards are offered to the player for cut.
Exploiting Shuffles 67
RESULTS
Figure 4-2 shows scatter plots” of the pre-shuffle vs. post-

shuffle card positions for four iterations of the two-pass
combo shuffle. A mathematically random shuffle would tend
to produce graphs with no apparent pattern. Inspection of the
plots reveals striking non-random patterns with diagonal
“streaks” indicating long sequences of cards that have
remained in the same relative order and in close proximity
through the shuffle.
Iteration # 1 Iteration # 2
OD ae ee,
<< at | ae" --
| ac
260 A = exer
ase eC 260 1 ones sar
ae
| —-
a wr ee ‘
208 ea 208 { age
es _ Be See
156 .- ae oe | 156 { aes --*
ee ni a |
AOA SEES eer . 104 | as . incom ED a
a TS icesrees
52 % Coie 52 i ce -
fee a at =
0 e a fo} ero aes? —
0 52 104 «156 «0208 «=2680S 312 0 52 104 156-208) 260 312
Iteration # 3 Iteration # 4
312 xs CO a— : oe 312 e
ae
. wenn?
280} yom & 260 | ne |
mass os -
208 |, COR ea 208 Soe os pee oeSe
inp ane . |
156 |- é ae e- | 156 re aah -
104f poe as is
ane Se mal
| ah Pr raanl
- Or iad * meron | 104 “7, SN asi |
Rant “4 Ps SS 4 |
52 ners B 52 e eon |
Sao deer
Bae era Peg ¥ |seo% er ate
0 te SPS 04 — oat
io) 52 104 156 208 260 312 ty) 52 104 186 208 260 312
Vertical y-axis = pre-shuffle position; horizontal x-axis = post-shuffle position.
FIGURE 4-2. PRE- VS. POST-SHUFFLE PLOTS
* [FS] p. 192.
SIMPLE TESTS FOR RANDOMNESS
° Test*
To test the randomness of the shuffling procedure a

combined two-way table of post-shuffle one-deck segments
against pre-shuffle one-deck segments was constructed. The
resulting counts are provided in Table 4-2:
TABLE 4-2. PRE- VS. POST-SHUFFLE CARD POSITIONS
Post-
Pre-Shuffle Position
Shuffle
Position
705-156
r32_| 6 peels aad esa
a0 | 38 [| 24 araco oS
eee dP Sa aie a eee
ae was a Sa ses
A
261-312 Same
A random shuffle would produce roughly equal counts (34
or 35 cards per cell) in each of the 36 cells, indicating each
card is equally likely to be found at any position in the post-
shuffle order. The counts in Table 4-2 are not uniform,
reflecting the non-random nature of this two-pass combo
shuffle. The observed chi-squaredt value y? was 234 at 25
degrees offreedom? (p < .01)8 allowing us to reject decisively
the null hypothesis** that this shuffle is random. Close
inspection of Table 4-2 shows some cells had a much higher
number of cards than expected. For example, a total of 71
cards from pre-shuffle positions 53 to 104 were found in post
shuffle positions 209 to 260—more than double the expected
* [FS] p. 166.
1 [FS] p. 146.
t [FS] pp. 156-157.
8 [FS] p. 193.
** [FS] p. 184.
number. Such cells show where cards tend to cluster in
predictable streaks—exploitable weaknesses in the shuffle.
Table 4-3 shows the shuffle data in Table 4-2 converted to
percentages. The two cells with bold borders indicate shuffle
weaknesses where the percentage of cards ending up in the
position was >30%:
TABLE 4-3. FOUR SHUFFLES BY HAND (%)
Pre-Shuffle Position
Position 105-156 57-208 | 209-260 261-312
1% 21%
27%
8%
105-156 14% 16%

157-208 16% 25%
CVShufjle was used to simulate four thousand iterations of

BJAP 3 on computer (Table 4-4):
TABLE 4-4. FOUR THOUSAND SHUFFLES BY COMPUTER
Post-Shuffle Pre-Shuffle Position ]

Position 53-104 | 105-156
| 157-208 | 209-260
4% | 9% | 20% | 23%
53-104 T% 16% 27% 22% 10%
105-156 17% 19% 19% 9% 17%
157-208 28% 13% 17% 8% 25%
17% 8% 15% 19%
26% 9 23% | 18%|
The software identified the same two shuffle weaknesses.

Variance” between the two sets of results was small. The
average variance was two percent.
* [FS] pp. 202-203.
Concordance among Shuffle Iterations
Kendall’s Coefficient of Concordance W can be used to

assess the similarity of outcomes for repeated iterations of a
shuffle procedure. ” Average post-shuffle positions for half-
deck segments were computed (see Appendix E) for all four
iterations and W was calculated using these half-deck
averages.
Measures of W range between 0 and 1, with values close to
1 indicating a high level of agreement. For this series of
shuffles W was 0.96 (p < 0.1) indicating a very high level of
similarity among repeated iterations of the shuffle. With this
level of agreement we can be sure the cards in pre-shuffle
positions 53—104 will end up in post-shuffle positions 209—
260 very consistently.
Differences in Average Post-Shuffle Positions
In a random shuffle the average post-shuffle positions for

each half-deck would be roughly equal. The Kruskal-Wallis”
test statistict H was computed using equation 4-1:
H=12/N(N+1)(¥R7/n) -3(N+1) 4-]
Where,
YR’ = squared sum of ranks of average post-shuffle positions
N=48
n=4
/4) - 3(48 + 1) = 44.52

H = 12/ 48(48 + 1) X (150154
The observed chi-squared value y? was 44.52 at 11
degrees of freedom (p <. Ol). The average post-shuffle
positions for half-decks in this shuffle were significantly
different from one another.
* [FS] p. 174.
+ [FS] pp. 200-201.
Exploiting Shuffles sp
MAPPING SHUFFLE WEAKNESSES
The above kitchen table analysis shows that this particular

shuffle is non-random, but how does this translate into action
at the blackjack tables? Let us begin by studying in more
detail the scatter plot of /teration # / in Figure 4-3:
312 cod FTI RAS ree? = ]
- Cn iceman e
oor * |
260 ey ¢
aad wee me
ee)
208 lo F Salat
7“ ¢ Kae
wer »
156 oes S cid oo
104 I kate Pe scan 8

noeese ee sa 4
- de er |
. or }
52 -
se". ee
o| Se ee Pe |
0 52 104 156 208 260 312
Vertical y-axis = pre-shuffle position; horizontal x-axis = post-shuffle position.
FIGURE 4-3. IDENTIFYING SHUFFLE WEAKNESSES
Look closely at the sequence of cards indicated by the

arrow. Tracing a line horizontally to the left (vertical) axis of
the graph tells us this sequence originated in positions ranging
from 50 to 62 in the pre-shuffle order. In a real blackjack
game, this corresponds approximately to the bottom thirteen
cards of the third half-deck from the base of the discard tray.
Furthermore, the horizontal (post-shuffle) axis of the graph
tells us these particular cards now occupy positions ranging
approximately from 200 to 250 in the post-shuffle stack. This
corresponds roughly to the second deck from the top after the
shuffle. In short, cards discarded early in the third half-deck of
the discard tray tend to end up in the second deck from the top
after the shuffle. Using pencil and paper, or a computer, you
will be able to quickly identify and map weaknesses in the
shuffle. It is enough to identify one or two reliable
weaknesses. The tracking method outlined later suggests
repeatedly locating one or more Aces occurring naturally
within these groups. Also, look first for different sequences
that end up in roughly the same post-shuffle positions. Close
inspection of Jteration # 1, for example, reveals a second
sequence of cards, in post-shuffle positions 200 to 250, that
originated in pre-shuffle positions 80 to 92. These cards start
in the lower part of the fourth half-deck in the discard tray and
consistently finish up in the second deck from the top in the
final stack. Tracking cards from such “converging” groups
will enable you to cut more than one sequence to the top if
you get the cut card. Figure 4-4 shows these two reliable
weaknesses in a shuffle map:
Bottom 13 cards >

Bottom 13 cards >
FIGURE 4-4. MAPPING SHUFFLE WEAKNESSES
Six decks of discards are stacked in the discard tray.

Starting from the bottom, i.e. the order they entered the
discard tray, cards are identified in half-decks as Segment A,
Segment B, Segment C, etc. Two decks remain in the cutoffs.
These cards are identified as half-deck segments I, J, K and L.
The cutoffs have been topped, i.e., placed on top of the
discards, to form the six-deck pre-shuffle stack. The arrows
indicate the exact positions of the two reliable weaknesses.
Based on our detailed statistical analysis, Aces falling
naturally within these 13-card segments will end up
consistently in the second deck from the top after the shuffle.
Note that some “streaks” of cards in the scatter plots
originated in the cutoffs. Since, in actual play, you know
nothing about the order of these cards, you will not be able to
“locate” them.
HOw RELIABLE ARE THE WEAKNESSES?
Let us put one “shuffle weakness” under the microscope to

determine how reliable it really is. Using the data from
Iteration #1, a simple regression analysis” was carried out for
the thirteen cards in pre-shuffle positions 50 to 62. The results
are shown in Table 4-5:
TABLE 4-5. A SIMPLE REGRESSION ANALYSIS
Pre-Shuffle Post-Shuffle 65
Position Position
60 |
50 201 =
51 203 |
52 208 i
553 214 ree a
54 215 Vertical y-axis = pre-shuffle position; horizontal x-axis =
55 219 post-shuffle position.
56 225
37 228 FIGURE 4-5. THE LINE OF BEST FIT
58 232
59 236
<b ane Fig. 4-5 shows that the post-shuffle
61 741 positions lie along a line of best fit
62 245 described by equation 4-2:t
ed Oy 4-2
Where,
x = post-shuffle position
a =y intercept
b= slope
y = pre-shuffle position
* [FS] p. 193.
i Forgotten Statistics shows this equation in the standard form Y = a + bX.
Using the statistical calculation software XLStatistics, the
values for a and b that gave the line of best fit for the data
were found to be: a= 12.154; b = 3.775.
Therefore, the straight line in Fig. 4-5 is described by:
el OAS PSY 4-4
Predicting post-shuffle positions using the average
The average post-shuffle position was 223.54. Using this

average to predict post-shuffle positions, the squared error
about the average SEjyg is given by equation 4-5:
n 4-5
Sle i Dx; Fy,eg
Where,
x; = any actual observation of a post-shuffle position
Xavg = a post-shuffle position as predicted by the average
SEG e200903
Predicting post-shuffle positions using line of best fit
Using the line of best fit (Eq. 4-4) to predict post-shuffle

positions, the squared error about the line SEjine is given by
equation 4-6:
a 4-6
SE tine Lx; fa
mS 2
Xline)
l=
Where,
x; = any actual observation of a post-shuffle position
Xline = a post-shuffle position as predicted by the line of best fit
SE tine = ]6
In this case, the line gives a much smaller error than the
average, making it a much more useful predictor of post-
shuffle position. The coefficient of determination an
measures how well the line fits the data:
y= l= eS 4-7.
SEs
Where,
SE line = Squared error about the line
SE qe = squared error about the average
wee 16 = 0.9938
2609.23
Measures of 7° range between 0 and 1, with values close to

1 indicating a very close fit. The observed 7° value (0.9938)
means equation 4-4 is very good for predicting post-shuffle
positions using pre-shuffle positions. In general, the thirteen
cards in pre-shuffle positions 50 to 62—the weakness in the
shuffle—end up in post-shuffle positions 201 to 245 very
reliably, time after time.
PREDICTING EXACT POSITIONS OF CARDS
Clearly, equation 4-4 could be used to make predictions of

the exact post-shuffle position of a card. For example, with
pre-shuffle position y = 50, Eq. 4-4 yields:
12194223 Xx 0 = 201
However, post-shuffle positions will vary due to chance.

Therefore, XLStatistics was used to determine a range of
* [FS] p. 189
probable values (95% prediction intervals) for post-shuffle
positions (Table 4-6):
TABLE 4-6. PREDICTION INTERVALS
Pre-Shuffle Lower Upper © |

Position Limit Limit 60 a
50 TY Se hae ie
il 202 208 S
52 206 PAUL
5
53 209-215 190 20 210 20 230 20 250 20
54 213 219 Vertical y-axis = pre-shuffle position; horizontal x-
55 217 222 axis = post-shuffle position.
56 221 226
57 224. «-239:~ST)s FIGURE 4-6. PREDICTION BANDS
58 228 234
59 P32 ae37 Figure 4-6 plots the prediction bands
60 236
241 ~~ for the post-shuffle positions. This
61 239
245 tells us, for example, that a card
62 243 249
starting in pre-shuffle position 50 ends
up in a post-shuffle position ranging from 198 to 204—95% of
the time! Unfortunately, the ordinal post-shuffle position of a
card is not easily obtainable in actual play.
However, simply knowing where this slug of thirteen cards
ends up after the shuffle is enough. Once this is known, we
focus on locating Aces in these highly predictable segments
using key cards. The segments are tracked visually through
the shuffle so that the Aces can be predicted accurately in
subsequent play.
THE GAMBLER KING
r Vhe two stunning blondes, one on each arm, immediately

reminded me of Frank Sinatra’s famous gag to a pit
boss at the Sands in Las Vegas. Flanked by two lovely
ladies, Frank had asked: ““What do you think of the cufflinks?”
“The Gambler King” was at least sixty, overweight, and
not particularly handsome. He was wearing sunglasses, a
white Armani suit, a white silk shirt fastened at the neck by a
Exploiting Shuffles a7
silver stud, and a pair of spotless white shoes. His long silver-
gray hair was pulled back tightly into a ponytail. His
“cufflinks,” about thirty-five years younger than him, had
never been inside a casino before. They were too frightened
to gamble, like fawns caught in car headlights.
I was expecting an Ace on the next hand so I changed up
1,000 French francs (about $200) and placed my bet on first
base. As I did, the King casually tossed 5,000 French francs
on to the blackjack table. The dealer quickly counted the
notes then, in one slick motion, lifted a stack of golden chips
from the tray; spread, counted, and restacked them, before
sliding them back across the green baize.
“All” snapped The King, pointing to first base. The dealer
pushed the King’s towering pile of chips alongside my single,
1,000-franc chip. As the dealer swished out the cards, the two
women started to giggle like schoolgirls. As predicted, the
first card out was the Ace, the dealer got a Five and, as our
second card hit the felt, they let out a little squeal. It was
another Ace!
Calmly, the King threw another 5,000 francs on to the
table, and, said: “Split.” There was a murmur of excitement.
A small crowd, sensing something unusual, started to gather
round. The dealer separated the two Aces, before arranging
another enormous stack of chips next to the second Ace, along
with another of my 1,000-franc chips. The Gambler King now
stood to win or lose 10,000 francs on this one hand.
The cards flashed again. Bang, bang! Two Jacks! Double
Vingt et un! The dealer drew a Ten and a Seven to bust with
22. The audience burst into spontaneous applause.
Waving theatrically to his gaggle of admirers, a little smirk
appeared on the King’s face that said: “It was nothing, really.”
As he hurriedly pocketed his 20,000 francs he winked at me
and whispered: “C'est assez pour ce soir’ (that’s enough for
tonight!).
LOCATING ACES
ACE LOCATION
The goal of the Ace tracker is to Jocate Aces during play,

visually follow the segment or segments containing the Aces
through the shuffle, and then cut them back into play. The
easiest way to locate Aces is to keep a tally of the actual
number of cards discarded. Aces that fall randomly in
segments previously identified as “weaknesses” become the
target cards.
Let us first map the movement of a single key card through
a shuffle. For example, half-deck A is niffled together with
half-deck B to form a single deck. A key card is at location 14
in A. One perfect riffle of the deck A+B, in initial order 1 to
52, produces the final order (reading across):
TABLE 5-1. MOVEMENTS OF A SINGLE KEY CARD
eel 25 oo 208 45) 30)

nS lemon 52) 7 33 FS 34
ORs See LU peOm wide S37 a 125 a38
13.39 Abe Se AlN 1G. AZ
17 43 18 44 19 45 20 46
21 47 22 48 23 49 24 «50
55 51-26 52
Locating Aces ik)

The key card is found at location 27 (alternatively, after
one modified perfect riffle, it can be found at location 28).
This suggests the rule: Only the bottom thirteen cards of any
half-deck should be used as key cards.
ACE SLUGS
Arnold Snyder defines a slug as: “ ... a segment which has

been identified as containing a large portion of either high
cards, low cards or aces.””° It is not practical to track every
Ace, or even every Ace slug, so the tracker should focus on a
small number of Aces observed during one or two rounds of
play. With a two-riffle shuffle, if you pick up a two-card
sequence anywhere within a 13-card slug, it will be washed
into 52 cards after the shuffle. Therefore, make sure you
pinpoint the sequence on the table to within six cards. The
location of a sequence can be pinpointed with any desired
precision by counting the number of cards dealt as they go
into the discard tray. This is easily done. Note, however that
you have to learn to count the cards in the order they go into
the discard tray. For example, busts and blackjacks may go
into the tray first. In some casinos, the dealer’s cards go in
first, followed by the players’ cards, while at others the
dealer’s cards go in last.
If you know, for example, the sequence is contained within
the slug from cards 50-55, you will know it will be within the
fourth half-deck from the top of the final stack. Alternatively,
if it is in the range 56—62 it will be in the third half-deck from
the top after the shuffle. This enables you to use a single key
card without running the risk of seeing too many false key
cards. As long as you are careful to pinpoint the sequence
accurately on the table, there is no change in the reliability of
the prediction about where it will be after the shuffle. In a six-
deck shoe the average distribution of Aces u in any segment
of cards is given by:
f=(A/ B)xc 52]
80 Locating Aces
Where,
A = number of cards in the segment
B= number of cards in the shoe
c = number of Aces in the shoe
For any quarter-deck: uw = (13 / 312) X 24=1 Ace. This

formula can be used to compute the average number of Aces
in any size segment in a six-deck shoe. For example, where
the dealer makes 39-card grabs uw = (39 / 312) X 24 =3 Aces.
The hypergeometric distribution™ was used to compute the
probabilities of Aces as shown in Table 5-2:
TABLE 5-2. PROBABILITY OF ACES

(in any Quarter-deck of a Six-deck Shoe)
Aces Prob. 1.00 |

0.90 |
0 0.346 0.80 |
0.70 |
0.391 0.60 |
0.50
0.195 0.40
Probability
0.30 |
0.057 0.20
0.10 |
0.011 (0100), (=
0.000 Aces
Total 1.000 FIGURE 5-1. PROBABILITY OF ACES

>/] 0.654 (in any Quarter-deck of a Six-deck Shoe)
Ace-poor segments (no Aces) occur about 35% of the time.

Nearly 40% of the time you get one Ace per quarter-deck.
Ace-rich segments (two or more Aces) occur about 25% of the
time. One or more Aces occur almost 66% of the time.
* TFS] pp. 168-169.
Locating Aces 81
Probability of Consecutive Aces
Given that there are, for example, two Aces in a quarter-

deck of a six-deck shoe, what is the probability Ps 4 that they
will be next to each other? This is given by:
Pall Pe
Where,
Pa. = the probability of two Aces being next to each other
P,"” = the number of permutations* of n objects taken r at a
time
Thus, the probability of two Aces being next to each other is:
Paa=1- 66/78 =0.154
Values of P4., for two through seven Aces are in Table 5-3:
TABLE 5-3. AT LEAST TWO CONSECUTIVE ACES
Aces
1.00
Prob. aa |
AWS: a |
30423 zoe |
4 0.706 Ecol |
5 0.902 * |
6 0.984 ao | | | |
7 0.999 : ; : :
Number of Aces in Quarterdeck
; }
FIGURE 5-2. AT LEAST TWO CONSECUTIVE ACES
(in any Quarter-deck of a Six-Deck Shoe)
As the number of Aces in a quarter-deck increases, so does

the probability that at least two Aces will occur consecutively.
For example, with seven Aces making up more than half of
the total cards in the quarter-deck, it is almost certain that at
least two Aces will occur consecutively.
* [FS] p. 186.
82 Locating Aces
A STEPLADDER SHUFFLE
In this section we show how to locate two Aces in the first

part of the BJAP 3 shuffle procedure from Chapter 4—the
stepladder shuffle. The effect of a stepladder procedure is to
move cards in the top half of any deck upwards through the
final stack while those in the bottom half remain undisturbed.
The shuffle definition is: six-deck, 66%

penetration, one-pass, four-pile, criss-cross, all
half-deck grabs, one-riffle, cutoffs topped.
This example shows two A&s observed in
different rounds of play. The first Aa@ is
located in half-deck C of the discard tray.
Within C, the A&@ is located in the bottom 13
cards. A second A4 is located in half-deck D.
Within D, this A@ is also located in the
bottom 13 cards.
The piles are broken into four 1'4-deck

piles. The quarters are identified as Pile 1,
Pile 2, Pile 3 and Pile 4. Dealer grabs F
and I. Cards are riffled together once.
Deck F+I placed on table to form final

stack. Dealer grabs C and the top half of the
already shuffled final stack F+I. Cards are
riffled together once.
Locating Aces 83
Deck C+(F+I), with A4 in bottom 26 ae
cards, added to final stack. Dealer grabs L er =
and the top half of C+(F+I). Cards riffled
once. Deck L+(C+F+I) added to final stack.
The stepladder procedure continues until all the cards have

been shuffled and placed on the final stack. In the final
shuffle map shown below, the letters grouped in brackets
represent a “grab” of roughly half a deck lifted from the deck
immediately below it in the final stack:
J+(A+G+D+K+B)
A+(G+D+K+B+H)
Final
84 Locating Aces
A TWO-PASS COMBO SHUFFLE
In this section the final stack of the above

stepladder shuffle becomes the pre-shuffle
stack in the second part of the BJAP 3
procedure—a simple riffle-and-restack shuffle
with the following definition: one-pass, two-
pile, all half-deck grabs, one-riffle, cutoffs
topped. Starting from the bottom, i.e. the
order they were placed on the final stack,
cards are re-identified in half-decks as
Segment A, Segment B, Segment C, etc. At
this point, we know only that one A&@ is shuffle
located in half-deck B and one A 4 is located in half-deck H.
The stack is broken into two 3-deck

piles. The top half is placed on the
table on the right side of the bottom
half (from the dealer’s point of view).
The halves are identified as Pile 1 and
Pile 2. Dealer grabs F and L. Cards
are riffled together once. Deck F+L
added to final stack. Dealer continues
to take half-deck grabs from both piles
until half-decks B and H, containing
tracked A 4s, are reached.
Aa H Aa B
Locating Aces 85
Dealer grabs B and H. Cards are riffled together once.
Deck B+H (with two A4s) added to final stack.
Final
Dealer grabs A and G. Cards are

riffled together once. Deck A+G added to
final stack. A player cuts.
Final
If you get the cut card, cut to keep as many tracked Aces as
possible in play. If possible, cut the Aces straight to the top of
the six decks, so they will be dealt in the first couple of rounds
of the next shoe. This gives you more time to /oad up other
key cards in the remainder of the shoe.
If other players cut the Aces out of play, try picking up
sequences from different tables. Load up early in a shoe to
give yourself plenty of time to monitor a second table while
waiting for the first to shuffle. Time your arrivals and
departures to observe the shuffles at both tables.
For analysis purposes the Aces in our example fell neatly
within the dealer’s half-deck grab size. In real games, dealers
86 Locating Aces
make grabs of different sizes, and alter their shuffles in
unpredictable ways.
Snyder warned: “If you don’t use your eyes at the tables,
but simply track shuffles based on numbers that work in maps,
you will be foiled again and again by plugs, breaks and grabs
that invalidate the map.” ”’
The diligent Ace tracker eyeballs a shuffle to make sure
Ace-rich segments are tracked to their actual final locations
rather than just where the map says they will be after the
shuffle.
THE GREEK SYNDICATE
(yi Nikolopoulos was fat and wore a wrinkled sports

jacket that reeked of cigar smoke.
The first sign that he knew we were card counters
came when he got the cut card. When the dealer turned the
six-deck stack on its side and offered it to him, George ran the
plastic card along its edge, pausing every deck or so to arch
his bushy eyebrows and tilt his big bull head questioningly at
us. When we shrugged, he grunted something in Greek,
convinced we were refusing him the secret of where the
“good” cards were.
His next ploy was to toss the cut card to us, but when he
placed an enormous pile of chips on third base and it vanished
in a blink, he decided it wasn’t such a good idea.
It was then that George came up with his “master” strategy.
Every time we placed a big bet, he ordered the dealer to place
an even bigger bet on our box, for him. Since we were
locating Aces, the results were much more to his liking, and he
had two or three big scores before the table closed for the
night.
When we arrived the next evening, George was already in
the casino. He had brought two friends. One of them looked
like his wife. Dressed all in black, with a big floppy hat and
sunglasses, she could have been Sophia Loren’s older sister.
Locating Aces 87
They seemed very happy to see us. They were playing
roulette and, apparently, they were losing.
As we placed our first bet at the blackjack table, there was
a frantic flurry of activity as “the Greek Syndicate” rushed
over from the roulette pit and started to pile up chips next to
ours.
In his haste George dropped a chip. It bounced across the
table and knocked down several carefully arranged stacks
belonging to the other players. The chips were all the colors
of the rainbow, and it took the dealer two or three minutes to
get them all back on to their correct boxes. Between us, and
the Greeks, there were one million drachmas (about $3,700)
riding on the next hand.
Finally, when the cards were dealt, the Ace fell on our box
and, by pure chance, a Queen followed it! Mrs. Nikolopoulos
let out an ear-splitting shriek, George roared, and their friend,
appearing to go weak at the knees, crossed himself and dabbed
his bald head with a hankie. George, with tears in his eyes,
shook hands with the other players, the dealer, the pit boss,
and us. He kept muttering over and over: “We're saved,
we're saved.” He gathered up his winnings, kissed his wife
on the cheek, and the three of them limped towards the door.
After they left, we asked the pit boss why George had been
crying. In broken English, he said: “Tomorrow ... the bank
... they take back George’s taverna. Tonight ... Tyche, the
Goddess of luck ... she smiles for George.”
88 Locating Aces
PREDICTING ACES
CARD PREDICTION
In his classic book The Theory of Gambling and Statistical

Logic (1977), Richard A. Epstein stated: “ ... because of
nonrandom shuffling, it is possible to predict with some
accuracy the arrangement of the deck, in which instance the
strategy encompasses the predictive function »78 and
“Obviously, the game expectation depends on the accuracy of
the predictive function. With perfect prediction, the
expectation reaches a value of close to 0.40.””"*
Three years before his book was released, Epstein
described a method of “predicting the succeeding card in a
sequence,”*” in a technical report written while he worked for
the Communications Division of the Hughes Aircraft
Company, in Los Angeles, California. Epstein’s ingenious
card predictive function depends on calculating the
probabilities of transition between packets of cards interleaved
during the shuffle.
* This 0.40 value is for perfect prediction of every card just prior to it being dealt.
Predicting Aces 89
THE PREDICTION PROBLEM
Consider again our simple four-card deck, all Spades, in

the initial order Ace, Two, Three, Four. In Fig. 6-1 the cards
are identified as c/, c2, c3, and c4, according to their location
in the initial order. Let the “A” set consist of c/ and c2, while
the “B” set consists of c3 and c4.
cl c2 3) c4
4 5 + 5 4 Ge +
4 4
q y e: o¢
A A B B
FIGURE 6-1. A SIMPLE FOUR-CARD DECK
Let the number of riffle shuffles n, = 1. The deck is cut

into two equal halves, with c/ and c2 in one half, and c3 and
c4 in the other, and the parts are riffled together once. Given
the first m cards of the newly shuffled deck (let m = 2), what
prediction can be made about the m + 1“ card?
ell c3 mt 1
q e
f+| ‘s
A B
FIGURE 6-2. THE PROBLEM OF PREDICTION
Epstein states: “Given the original sequence and the first m

cards of the final sequence, the (m + 1°) card must be one of
2” cards, n shuffles having transformed one sequence into the
other.”*' The third card must be one of two possible cards. It
is) either’ c2, from the A set; or c4; from the B set. The
probability of the next card coming from either set is 0.5.
Dr. David Bayer, of Columbia University, New York, and
Dr. Persi Diaconis, of Stanford University, California,
provided a formula (Eq. 6-1) for calculating this probability in
the academic paper Trailing the Dovetail Shuffle to its Lair:
90 Predicting Aces
PA, 8B A ?
(A+B) 6-1
They wrote: . cards drop from the left or right heaps

“eo
with probability proportional to the number of cards in each

heap. Thus, if there are 4 and B cards remaining in the left
and right heaps, then the chance that the next card will drop
from the left heap is 4 / (4 + B).”*
For a 52-card deck, the probability p, of the first card
falling from set A = 26 / (26 + 26) = 0.5. The probability pz
of the first card falling from set B = 1 - py = 0.5. These
probabilities apply to the second and subsequent cards of both
halves, until the riffle is complete. It is possible at any point
in a riffle that not one, but two cards may fall from either half.
However, this happens with a much smaller probability (0.25).
David Aldous, of the University of California, Berkeley,
with Diaconis, remarked: “The model produces shuffles with
single cards being dropped about 2 of the time, pairs of cards
being dropped about % of the time, and i cards blocks being
dropped about 4' of the time.’ Also, if two cards fall from
set A, and none from set B, the probability that the next card
will fall from set B increases: pg = 1 - 0.48 = 0.52. This
prevents an excessive number of cards dropping from one
heap or the other.
Applying the formula to the problem as shown in Fig. 6-2,
it is equally likely that the third card will come from set A or
B:
A 1
Dee tANB) =) (+l) = |
Nd
The probability of a correct prediction is the same as

guessing heads or tails by tossing a coin.
APPLYING EPSTEIN’S PREDICTION THEOREM
Here we apply Epstein’s theory of predicting specific cards

(Aces) to a blackjack game. Suppose you see cards laid out
like this:
Predicting Aces 91
Key |
Ace a
BOX3 BOX2 BOX1
FIGURE 6-3. MEMORIZING THE KEY CARD
Take careful note of the card that will lie underneath the
Ace when the dealer “scoops” up the cards. This is the key or
signature card. The Ace is the target card. After the dealer
pays off the winning bets, he scoops up all the cards,
preserving their original order. This is done so that hands can
be reconstructed from the discard tray if there is a dispute
later. Note, on box 3, the first card goes under the second
card, which goes under the third card. These three cards go
under the two cards on box 2, which go under the two cards on
box 1. Finally, the packet of players’ cards are used to scoop
up the dealer’s cards, before all cards are turned over and
placed face down in the discard tray. The key card is now on
top of the Ace in the discard tray. If we could “freeze” the
game at this point and look the cards in the discard tray, they
would look like this:
ESD) c36 (esi c3 c4 42 c43
CE) GEER
Ser esi
FIGURE 6-4. A SEQUENCE IN ITS INITIAL ORDER
(in the discard tray)
In Fig. 6-4, the cards are identified as c35, c36, c37, c38,
c39, c40, c41, c42, c43 and c44, according to their location in
the initial order.
Each round is dealt in a similar fashion until the cut card
comes out and the cards are shuffled. If we could look at the
92 Predicting Aces
section of the original cards just prior to the Ace we might see
something like this:
cl c27 ol3) G2 c39 c40 c28 cl4 Glo) Os}
FIGURE 6-5. THE KEY CARD IN THE FINAL ORDER

(after a two-riffle shuffle)
In Fig. 6-5, the cards are identified as c/, c27, c13, c2, c39,
e40.c/5,¢14, el) and c3. Wet the Ce.set consist.of.cl,.c2,
eater) seb consist 0kc?/7,.c25, the Eset cl3,.cl4, cl 5,
and the “F” set c39, c40. The new shoe is offered for the cut.
You cut, everyone bets and the game begins again. As the
cards are dealt wait for the key card (c40 from F) to appear as
shown in Fig. 6-6:
i
aca
BOX 3 BOX2 BOX 1
FIGURE 6-6. THE KEY CARD RE-EMERGES

(during the first round of the next shoe)
Watch closely for the target card, the Ace (c4/ from F), a
few cards later. Four more cards are dealt but the Ace does
not appear! The round is complete. Since we have seen the
first ten cards of the newly shuffled shoe, m = 10. What
prediction can be made about the m + 1* card? Since n =2 the
eleventh card must be one of 2” = 2” or four possible cards. It
is either c4/ from F, c29 from D, c/6 from E, or c4 from C.
Predicting Aces 93
THE PREDICTION PROBLEM SOLVED
It is not possible for a human to discern easily between

multiple sequences of interleaved cards, but with some
simplification, it is possible to apply a single-sequence version
of Epstein’s method. By observing three-card patterns in the
cards that have been dealt, Epstein showed how to calculate
precisely the probability that the next card will come from any
one of the four sets.** The four observable patterns are:
TABLE 6-1. THE PATTERNS OF PREDICTION
AAA AAB- ABA _ ABB
For example, when we see the pattern AAA, the probability

of a transition to Set B (AAAB) equals | since the probability
of no transition (AAAA) = 0.
Similarly, when we see AAB the probability of a transition
to A (AABA) =
r, or 8/9 and the probability of no transition (AABB) =
] - r, or 1/9
ABA has the same probabilities as AAB. Lastly, for ABB,

the probability of a transition to A (ABBA) =
r2/ 1-7, = (8/81)/ (1 - 8/9) or 8/9
and the probability of no transition (ABBB) =
r3/1-r,=(1/81)/(1 - 8/9) or 1/9
Applying these formulas to the current problem, we test

whether the next card is more likely to come from sets CD or
EF. In Fig. 6-5 above, the last three cards dealt from CD and
EF form the pattern EEC, which is the equivalent of AAB.
The probability of a transition to set E (EECE) =
94 Predicting Aces
r, OF 8/9
and the probability of no transition (EECC) =
1 -r,or 1/9
Therefore the next card is more likely, with probability 8/9,

to come from EF than CD.
Testing whether the next card is more likely to come from
E or F, we note in Fig. 6-5 that the last three cards dealt from
E and F form the pattern FEE, which is the equivalent of
ABB. The probability of a transition to F (FEEF) =
r2/ 1-7, =(8/81)/(1 - 8/9) or 8/9
and the probability of no transition (FEEE) =
r3/ 1 -r,=(1/81)/ (1 — 8/9) or 1/9
Therefore the next card is more likely, with probability 8/9,

to come from F than E, and is most likely, with probability 8/9
X 8/9 or 64/81, to be c4/ from F. We predict the Ace will be
dealt as the first card on the next round:
mM
ie
BOX 3 BOX2 Box 1
FIGURE 6-7. THE TARGET ACE APPEARS

(on the following round)
The Ace falls on box 1! We are predicting precisely where

Aces will appear in the game! To forecast where an Ace will
Predicting Aces 95
appear, all you have to do is remember the key card that lies
on fop of it when it goes to the discard tray.
A PREDICTION RULE
Recall from Chapter 3 that, with a two-riffle shuffle, three

cards in the final order most frequently separate two cards that
were originally neighbors in the initial order. This suggests
the rule: subtract the number of cards that come out after the
key card from four and bet on that number box. For example,
if two cards come out after the key card, 4 - 2 = 2, bet on box
oe
Note that, in the example above, four cards came out after
the appearance of the key card (Fig. 6-6). Applying the rule,
we subtract four from four to predict the Ace will fall on box
zero! Since there is no box zero, a player might be tempted to
bet on box |. However, our shuffling experiment shows that,
in this situation, the Ace will fall on the predicted box 15
times on average for every 100 predictions made. This is too
few to make the game profitable under most circumstances.
Do not bet. If no cards come out after the appearance of the
key card, subtract zero from four to predict the Ace will fall on
box 4. If only three betting boxes are occupied in the next
round the Ace will fall as the dealer’s first card. Avoid this by
covering more, or fewer, than three boxes. For example, with
two boxes covered, the Ace would fall as the second card on
box 1. It does not matter if the Ace is first or second card.
Alternatively, you may cover four boxes to catch the Ace on
box 4, but if the dealer gets the Ace it costs you, on average,
34 to 37%. You may cover five boxes to make this less likely.
If the Ace does not fall on the predicted box play the hand
using basic strategy.
HOW TO MEMORIZE KEY CARDS
Look for the first Ace that appears. Focus on memorizing

the key card for that Ace and make sure you get it firmly fixed
96 Predicting Aces
in your mind. Take things slowly. Always use a key card that
will be immediately underneath the Ace as the cards are
scooped up at the end of the round. Sometimes the key card
and the Ace both come out on the same round. Try
memorizing a single key card for two Aces lying next to each
other. Use the dealt Ace as the key card for the Ace yet to be
dealt. Always look for sequences of two or more Aces
separated by a Ten-value card (10, J, Q, or K). Memorize the
card before the first Ace and the card separating the Aces but
not the Aces themselves. If the first Ace and the Ten-value
card come out, you know the second Ace is close behind.
Some key cards will be cut away during the shuffle, so you
must learn to hold more than one key card per shoe. The
optimum way to locate Aces is to learn to hold four or more
key cards in your memory. A card memory system must be
capable of permitting easy, immediate recall of key cards as
they are dealt. Some players use words to represent key cards.
Others allocate images to the cards. Short-term memory’s
limited digits span of seven, plus or minus two can be
overcome by using both types of system together. With
practice you can rapidly memorize several key cards using
such systems.
MEMORIZING THE TARGET ACE
Epstein (1977) pointed out that, if asmall group of cards is

known to contain an Ace, simply knowing its suit increases
the probability of one or more additional random Aces in the
group.
If, for example, in a two-riffle game, you see a key card
dealt as last card of a round, with no cards following it, you
know that, on average, the next four cards will contain an Ace.
The probability P that these four cards will include at least
one more additional random Ace was calculated for the one-
deck case as follows:
[P= Ce Gr Ge = (1.0607
CONC i Cove a Cree
Predicting Aces 97
Where,
P = the probability of one or more additional Aces given an
Ace
C,” = the number of combinations* of n objects chosen r at a
time
If you also know the Ace suit, the probability P’ that the
four cards contain at least one more Ace, goes up:
P= ACENGAELG HGH GX Che 01094

ey
Where,
P' = the probability of one or more additional Aces given an
Ace of known suit
C,”= the number of combinations ofn objects chosen r at a
time
Epstein concluded: “... knowledge of the specific Ace has
oe
increased the probability of one or more additional Aces in the

sample.”*
Lastly, the probability P” of additional Aces when the Ace
suit is known was calculated for the six-deck case using the
hypergeometric distribution:
TABLE 6-2. ADDITIONAL ACES FOR SIX DECKS
Aces Prob. 1.00

0.90
0 0.79 0.80
=> 0.70 |
1 0.19 = 0.60 |
© 0.50 |
2 0.02 8 0.40 |
a. 0.30 |
3} 0.00 0.20 |
0.10 |
Total 1.00 0.00 |
>/ 0.21 0 1 2 3
Additional Aces
FIGURE 6-8. ADDITIONAL ACES FOR SIX DECKS
* [FS] pp. 148-149.

98 Predicting Aces
In moving from one to six decks, the probability of one or
more additional Aces has increased from 0.1694 to 0.2064.
From a practical standpoint, if a player can pinpoint an Ace
to within a few cards, it pays to know its suit. Memorizing the
target Ace, as well as the key card that precedes it, will
increase your chances of receiving additional random Aces
within the same few cards.
STEERING THE ACES
In years gone by, players would spread their bet across

multiple boxes to steer the Ace to the money. By carefully
calculating the number of boxes needed, players could “‘catch”
the Ace as their own first card, force a “bad” card (4, 5 or 6) to
the dealer,”° or increase their chances of getting a Ten-value as
their second card. Such tactics, when successful, led to
substantially increased advantages for the player, even when
the “hit rate’ was much less than 100%. In most of today’s
games it is difficult to steer cards with such accuracy. When it
goes wrong, for example, if a player accidentally steers a Five
to his own hand, it can be very costly. Also, Snyder (2003)
has warned that the “steering hands,” i.e. those hands that do
not catch the Ace, can cost the player up to 2% per hand in
negative decks. He advised: “players who sequence aces
would often find it more profitable in the long run to play a
single hand when the key card(s) predict an ace is coming, not
multiple hands in an attempt to catch it.”*’ Situations still
arise, however, where it can pay to deviate from basic
strategy. For example, suppose you see the cards in Fig. 6-9:
Predicting Aces O95

q a6 ce +
| * sy * ao *
a He ipl pekatse
ee SEA A
BOXx3 BOX 2 Box 1
FIGURE 6-9. STEERING THE ACES
According to basic strategy, the player at box 2 should

stand with hard 15 against dealer up card Four. However, an
Ace tracker may deliberately hit the hand until it busts so that,
at the end of the round, the Ace of Clubs and the Ace of
Hearts are scooped up and placed together in the discard tray,
with the Seven of Clubs above them as the key card. With a
minimum bet, the cost of the “wrong” basic strategy play is
relatively small (about 0.20 of the bet, for this example)
compared to the large gain that will be made later when the
Aces hit the big money. Note also, that if you memorize a key
card in a hand next to an Ace (for example, the Five of Spades
in Figure 6-9) and the hand busts, the key card goes to the
discard tray without the Ace. Conversely, if the hand with the
Ace busts, the Ace goes to the discard tray without the key
card. Sometimes, an Ace will be on top of the discards and
more discards will be placed on top of it. You can use a
discard as a key card if you are sure it is placed immediately
on top of the Ace.
FALSE KEY CARDS
A false key card is one of identical suit and value to a

genuine key card but with no Ace following. Since any false
key card will not fool us occurring after a real key/target card
sequence, we need only concern ourselves with one or more
false key cards that appear in isolation and before the genuine
100 Predicting Aces

key card. /f you observe two identical key cards next to each
other, or separated by one or more cards, do not bet. The
equation for the hypergeometric distribution:
P(X= i) =h(i, n, M, N) 6-2
was used to calculate the probability P;-; of one false key card
occurring in isolation before the real key card within 26 cards.
Since half the time, on average, the false key card will occur
after the genuine sequence, the size of the sample n is assumed
to be 12 cards (26 cards minus the genuine two-card sequence,
divided by two). The number of successes in the population
M is 5 (there are five duplicates of the real key card in a six-
deck shoe) and the population size N is 310 (the number of
cards in six-decks minus the two cards in the genuine
sequence):
Piz; = hd, 12, 5,310) =0.17
And, for all possible values of P(i):
TABLE 6-3. PROBABILITY OF A FALSE KEY CARD
I AND 1.60} |
0.90 |
f. 032 0.80
ce 0.70 f
| |
1 O77 = 0.60 | |
@ 0.50 | |
2 0.01 ee
2 0.40 |
3-02.00 0.20
0.10
| || |
4 0,00 0.00 0 1 2
: 3 4
%
5
5 0.00 False Key Cards
Total 1.00
>/ 0.18 FIGURE 6-10. PROBABILITY OF A FALSE KEY CARD
Summing for one or more false key cards: P(i) = 0.17 +

0.01 = 0.18. Since we are tracking four key cards per shoe,
we must also calculate the probability Pj) of none, one or
more different key cards being duplicated. The probability
P(1) of one key card being duplicated is 0.18. The probability
of two key cards being duplicated P(2) is 0.18 X 0.18 = 0.03,
Predicting Aces 101

(P(3) and P(4) = 0). Finally, 1 - (P(1) + P(2) + PB) PA)
yields the probability P(0) for zero key cards:
TABLE 6-4. PROBABILITY OF FALSE KEY CARDS
j AND 1.00 |
0 0.79 cal
1 0.18 Sal
2 |
0.03 ao | |
3 0.00 std |
4 0.00 oa |
Total 1.00 : False zsCards
=f 021
FIGURE 6-11. PROBABILITY OF FALSE KEY CARDS
For one or more different key cards P(j) = 0.18 + 0.03 = 0.21.
POINTER CARDS
Another subtle principle from card magic, pointer cards,

can reduce the number of bets made on false key cards. Note
that 22 cards. in’ every deck (Ace, 3; 5,6, 7, 8; 9-of Clubs,
Spades, Hearts, and 7 of Diamonds) have an asymmetric face
design:
FIGURE 6-12. ASYMMETRIC FACE DESIGNS
In Fig. 6-12, for example, the genuine key card (left) has
the majority of its pips facing up; therefore a duplicate card
would not fool us with the majority of pips facing down.
If you examine Bee® casino decks closely you will see
(Fig. 6-13) that every card has a small but discernable
difference in the width of the white space between the corner
pips and the edge of the card. A sharp-eyed sequence tracker
can utilize this, also. Memorizing a card and its alignment
102 Predicting Aces

halves the number of bets made because of false key cards.
P(j) is reduced to ~0.10.
im
Wide Space>| ¢ 4
+ q © Narrow Space
FIGURE 6-13. ASYMMETRIC WHITE SPACE
Asymmetric Back Designs
Steve Forte, an expert in gambling /egerdemain, offers this

method to “read” Aces from the back: “the vast majority of
decks are cut slightly off-center. At first glance the backs all
look the same, but when you carefully examine the edges, the
small ‘triangles’ around the sides vary dramatically in size.”**
Figure 6-14 shows an example using the back of an Ace
from a new, standard 52-card Bee® deck (No. 92 Club
Special, Match Blue Diamond Back No. 67):
\x
KOK.
RS
+ XXYY4°4°OO)
KKK
WY
+ XY
x
tatatatatets
$,
*
OK
+4 + + +
XXX
¢ AY" %* %y)ne% ’ %
+ ry + afore
+ ny x)PKS wsDX)fe
ate’ 4
4,
FIGURE 6-14. ASYMMETRIC BACK DESIGN
The card has big triangles running along the top edge and
little triangles along the bottom edge (this edge is visible when
the card is about to be dealt from a dealing shoe). Trackers
can also use this to predict whether their first card, the dealer’s
hole card, or their own draw card, will be “high” or “low.”
Playing the turn applies to hand-held games only. Forte
explained: “As he receives his cards, the advantage player
simply turns them to the desired alignment. Eventually, after
a few deals, he’s got the deck ‘marked.’ All the little cards
have the little triangles on top; all the big ones have big
Predicting Aces 103
R
triangles on top.”*’ One very advanced play for shoe-dealt
games involves noting the back-alignment of any Aces
tracked. If an Ace is due on first base the player checks the
alignment of the next card to be dealt. If it matches that of the
tracked Ace he bets on box 1. If not, he bets on box 2.
Reading the Edges
Here is Laurent Bujold’s secret for cutting Aces. Note how

the diamond pattern on the back of a Bee® card is “bled off”
at the edges. This causes stripes along the sides of the deck as
shown in Figure 6-15:
FIGURE 6-15. “READING” THE EDGES OF A BEE® DECK
Look closely and you will see one card (reversed in

direction) with edge markings out of line with the others. A
row of dashes is visible in the white spaces between the
stripes. Only one such readable Ace is required to gain a huge
advantage.
PUTTING IT ALL TOGETHER
Predicting Aces at the table is easier than the theory makes

it look. Here we show how to predict an Ace in a game with
the two-pass, combo shuffle as described in Chapter 4.
In this example, you have already seen the first four rounds
of play. Having counted the cards as they were dealt, you
know there are exactly 49 cards in the discard tray. Therefore,
most of the cards in the next round will fall into the
“weakness” segment identified in Chapter 4.
104 Predicting Aces

According to Table 5-2, two out of three times (0.654) the
cards will include at least one Ace. If two Aces are dealt, the
probability they will be next to each other is 0.154 (Table 5-3).
On the fifth round, an Ace appears:
HT
fober[4
BOXx4 BOXx3 BOX 2 BOXx1
FIGURE 6-16. AN ACE IN A TRACKABLE POSITION
You note the key card that will be under the Ace when the
dealer scoops up the cards.
You know the sequence lies within a six-card slug ranging
from positions 56—61 in the discard tray. From Chapter 4, you
know this sequence is very likely to be in the third half-deck
from the top after the shuffle. However, to make sure, you
eyeball the shuffle closely, and track the sequence to its final
location.
You are offered the cut card, and cut to bring the Ace
straight to the top of the six decks.
On the first round of the new shoe ...
i"
Oe oe | Flee
BOX 4 BOX 3 BOX2 BOX1
FIGURE 6-17. A FALSE KEY CARD APPEARS
Predicting Aces 105

... a key card comes out, but you realize it is false because its
pips are facing the opposite way to those of the real key card!
A few cards later, the genuine key card appears. You count the
number of cards following the real key card (two) and predict
the Ace will appear on box 4 - 2 = 2 in the next round. You
make a big bet on box 2...
in
ee
BOXx4 BOXx3 BOX 2 BOX 1
FIGURE 6-18. THE ACE FALLS ON THE PREDICTED BOX
... the Ace falls on box 2! This is an extremely elegant and

incredibly powerful technique.
THE CURSE OF WANG
‘ J ang was an old Chinese guy who could be found

most evenings sitting at third base. All the dealers
called him the “Zen Master” because win or lose, he
never lost his temper and, regardless of whether he
got good cards or bad, he never complained.
However, Wang was very superstitious. He kept a little
spider in a matchbox. Whenever he was dealt a stiff he would
peek inside the box to consult the spider. Depending on which
106 Predicting Aces

side of the box the insect was sitting on, Wang would hit or
stand.
One night Wang was dealt a Two and a Ten for a hard total
of 12, while the dealer’s card was a Three. The old man
looked inside the box and the position of the spider told him
he should stand.
Wang stood. The dealer drew a Seven and a Ten for a total
of 20 and all five players at the table lost. It was too much for
one high roller from New York. He stood up and yelled at
Wang: “Learn how to play the game, buddy!” before storming
off. Wang never even blinked.
On the next round Wang got a Four and a Queen for a total
of 14. Again, the spider signaled: “stand.” The dealer got 19
and everyone lost again.
A man and his wife were completely disgusted and got up
to leave. The husband pointed angrily at Wang and said to the
dealer: “That old fool has put a curse on the whole table!”
Wang didn’t flinch.
The old man’s play was so pathetic even the pit boss was
shaking his head. Eventually, he said to the dealer: “Let him
play any darned fool way he pleases,” before wandering off.
I was catching plenty of Aces and, with no pit bosses
watching, betting the table maximum without any heat
whatsoever. The winnings were piling up fast.
After a while, Wang quit. Since I’d filled my boots I
decided to leave too. Outside, I noticed Wang lurking in the
dark alleyway across the street.
I walked over and whispered: “Wang, here’s your hundred
dollars. See you again tomorrow.” Without saying a word,
the “Zen Master” took his share of the winnings and
disappeared into the night.
Predicting Aces 107

7
MATHEMATICAL EXPECTATION
BEAT THE DEALER
In his best-selling blackjack book

Beat the Dealer, Edward Thorp wrote:
“Suppose there are five cards left,
mostly Aces and Tens, and that you
decide to take five hands. Then you
get all five of the cards in this favored
group, and the dealer gets none of
them, for he runs out of cards and must
shuffle before dealing the first card to
himself. If you now get a Ten as your first card this gives you
a 15 to 20 per cent advantage; starting with an Ace gives youa
35 to 40 per cent advantage.””
VALUE OF AN ACE AS FIRST CARD
In Casino Holiday (1970), Jacques Noir gave +51% as the

player’s percentage with Ace as first card. He noted: “Thorp
gives an estimate of this number as 35-40%. The 5% spread
in his estimate indicates it is just a crude guess, and this should
not be interpreted as a real difference of opinion.”
Griffin (1981) stated: “the player’s expectation as a
function of his own initial card [is 52% for an Ace].””* Snyder
108 Mathematical Expectation

(2003) observed: “Depending on the number of decks in play
and, especially, on the rules, the ace may be worth anywhere
from about 51% to 53% to the player’s hand.”” Julian
Braun’s blackjack simulation data” enable us to produce
Table 7-1, the player’s precise expected value E(X)* with Ace
as first card, for one and four decks:
TABLE 7-1. EXPECTATION WITH ACE AS FIRST CARD
Ace 1-deck 4-decks

2 -0.005 -0.022
3 -0.031 -0.050
4 -0.069 -0.077
5 -0.108 -0.103
6 -0.057 -0.074
T +0.052 +0.038
8 +0.276 +0.270
o +0.594 +0.582
10 +1.445 +1.433
J +1.445 +1.433
Q +1.445 +1.433
K +1.445 +1.433
A +0.406 +0.332
Totals +6.84/13 +6.63/13
E(X) = +0.526 = +0.510
Expected value for all first cards for one and four decks are
shown in Table 7-2:
TABLE 7-2. EXPECTATION FOR ALL FIRST CARDS
FirstCard A 2 3 4 5 6 i 8 y) 10
One Deck +0.53 -0.13 -0.15 -0.16 -0.21 -0.20 -0.18 -0.08 0.00 +0.15
.-043, -0.15.-017 -021 -0.21. 0.18. -0.09.. -0.01 +0.14

Four Decks +051
With an Ace as first card, expectation ranges from +0.51 to
* TFS] p. 161.
Mathematical Expectation 109

+0.53. Rules giving the player more options when holding an
Ace, such as soft doubling, re-splitting Aces, and doubling
after splits, increase the value of the Ace to the player.
If a hand with Ace as first card is played many thousands
of times the expectation in a four-deck game is +0.51
regardless of the value of your second card. However,
expectation for any individual hand depends on the value of
the card that falls on the Ace. There are equal numbers of
each second card in neutral decks. There are slightly less
Aces because one is used up as first card. “Good” second
cards for the player are 7, 8, 9, 10, J, Q, K, and Ace. “Bad”
second cards are 2, 3, 4,5, and 6. Note: Ifa 2-6 value falls on
an Ace, expectation is negative!
For a detailed analysis of how to play the hand if the player
does not receive a Ten-value as second card on the Ace, see
the article “More on the Ace in Hand ...” by James Grosjean
and Previn Mankodi.”°
Probability of a Ten as Second Card
Eq. 7-1 gives the conditional probability of receiving a

Ten-value as second card, given that the first card is an Ace:
P(T) |p(A) = p(A) X p(T) /p(A) Fel

For six decks:
P(Z) |p(A) = (24/312 K 96/311) / (24/312) =0.31
To increase the chance of getting a natural keep a running

count of each round individually and key Aces only in rounds
that have a negative count.
* [FS] p. 149.

HOW OFTEN DOES THE ACE “HIT THE MONEY”?
Player's Expectation—100% Correct Predictions
If the Ace landed on the predicted betting box every time,

the player’s expectation would be +0.51. Each one hundred-
unit bet would earn 51 units! Unfortunately this is unrealistic.
The exact number of Aces that will hit the money can vary
greatly, depending on the number of cuts and riffles in the
shuffle.
Player’s Expectation—38% Correct Predictions
According to the two-riffle shuffling results reported

earlier, 38% is a more realistic estimate for predicting the
correct box. The Ace will land on the expected box in 38 out
of every one hundred predictions made:
@ 100% expectation = +0.51

@ 38% expectation = 0.38 X +0.51 = +0.1938
Recall from Chapters 3 and 6 that broken sequences and

false key cards further reduce the player’s expectation. In
Chapter 3 the probability of broken sequences in a simple
riffle-and-restack shuffle was estimated to be 0.15. In other
words, fifteen times in every one hundred predictions the Ace
will fail to land where expected because it has been cut away
from the key card in the shuffle.
In Chapter 6, the probability of false key cards was
calculated as 0.10. Thus, false key cards will fool us into
betting on phantom Aces a further ten times. This means we
must reduce the 38% hit rate by a further 25% to leave 13%.
Lastly, since the Ace hits the predicted box only 13 times
in every hundred bets, we must deduct the cost of playing the
other 87 hands using basic strategy.

SKILLED VERSUS UNSKILLED PREDICTORS
A useful way to view “hit rate” is to compare the expected

results of a skilled predictor, such as a professional Ace
tracker, with an unskilled player—that is, someone who
predicts Aces merely by guessing. For instance, in our Six-
deck game, the unskilled “guesser” finds that, on average, he
predicts successfully one Ace in every thirteen attempts: =
13 X 24/ (24 + 288) = 1, yielding an overall hit rate of 0.07.
The skilled Ace tracker, on the other hand, predicts an average
of almost two Aces in thirteen attempts: p = 13 X 42 / (24 +
288) = 1.75, with a hit rate of around 0.13. In other words, the
skilled player is nearly twice as good at predicting Aces as the
unskilled player (Table 7-3):
TABLE 7-3. SKILLED VS. UNSKILLED PREDICTORS
Aces Skilled Unskilled eed aes ees

0 0.15 0.35 >, ees
0.70 } | t|e nie)
i} 0.31 0.39 = 0.60 |
9g 0.50
2 0.29 0.19 © 0.40 |
Eaves) |
3 0.17 0.06 0.20 |
0.10
4 0.06 0.01 0.00
5 0.02 0.00 0 eV eeee ee
No. of Aces predicted successfully
6 0.00 0.00 in 13 attempts
Totals 1.00 1.00 FIGURE 7-1. SKILLED VS. UNSKILLED

>/ 0.85 0.65 PREDICTORS
Table 7-3 reveals the skilled predictor makes one or more

correct predictions (for every thirteen attempts) 85% of the
time while the guesser manages the same only 65% of the
time. However, consider how slight the difference is between
the expert and the guesser. The expert predicts just six more
Aces in every hundred than the guesser but, because the Ace is
the most important card in blackjack, these six hands give the
expert a significant advantage over the house—the source of
the expert’s lucrative profits.

WHEN THE DEALER GETS THE ACE
The player expects to get the Ace “on purpose” 13 times in

every one hundred attempts. This will be true at a full table,
where the dealer has little chance of getting the Ace “by
accident.” However, at less than full tables, the dealer will
sometimes get the Ace accidentally instead of the player,
further reducing the expectation of the player.
The probability of the dealer getting the Acecan be
calculated fairly precisely, given the number of riffles, the
dealer’s riffle “signature” and the number of betting boxes
covered. But, for a more general “formula,” we need a more
general rule.
Snyder (2003) proposed a “rule of thumb” solution that
says, in the long run, the player and the dealer will share the
Aces 50/50.”° This works for any number of riffles and any
number of boxes covered. It is a very conservative rule, since
you are effectively saying the dealer will get one tracked Ace
by accident for every extra Ace the player catches using skill,
but it has the great virtue of simplicity.
When the dealer gets the Ace, it costs the player 0.34 with
multiple decks (0.37 if the Ace falls as the dealer’s hole card),
but this is more than compensated for by the fact that, when
the player gets the Ace, it earns +0.51 per bet. The player's
net expectation is +0.51 - 0.34 = +0.17. However, this is
before any deductions are made for broken sequences, false
key cards or basic strategy disadvantage.
MATHEMATICAL EXPECTATION
The author’s formulas, given below, can be used to

calculate mathematical expectation E(X) when predicting
Aces.
Eq. 7-2 can be used to calculate the probability P(h) that
the Ace will Ait the money:
P(h)=a-(b+f) 7-2
Where,

a = probability of an average number of cards separating
initially adjacent cards
b = probability of broken sequences
f= probability of false key cards
Assuming a = 0.38, b= 0.15,f= 0.10
P(h) = 0.38 - (0.15 + 0.10) = 0.13
Therefore, the probability P(m) that the Ace will miss the
money = | - A = 0.87. At this point we invoke Snyder’s rule
of thumb—for every additional Ace the player successfully
predicts using skill, the dealer gets an Ace by accident. In this
case that means the dealer gets six additional Aces. P(m) is
reduced from 0.87 to 0.81, while P(d), the probability of the
dealer getting the Ace by accident, becomes 0.06. Eq. 7-3 can
now be used to calculate mathematical expectation E(X):
E(X) = E,h + E,d + E3m 7-3
Where,
E = player’s expectation if the Ace hits the money
E, = player’s expectation if the dealer gets the Ace by accident
E; = player’s expectation if the Ace misses the money
h = probability that the Ace will hit the money
d = probability that the dealer will get the Ace by accident
m = probability that the Ace will miss the money
Assuming £, = +0.51, E> = -0.34, E; = -0.005 h = 0.13, d=

0.06 m= 0.81
E(X) = (0.51 X 0.13) + (-0.34 X 0.06) + (-0.005 X 0.81)

= 0.0663 - 0.0204 - 0.0040
= +0.042
Against the two-riffle R&R shuffle, Ace Prediction
achieves single-trial win probability p ~0.52 X (1 - 0.083)” =
0.48, loss probability gq~0.48 X (1 - 0.083) ~0.44.°% For

simplicity, the increase or decrease in fortune for each
play is
assumed to be one unit. Under these assumptions, a = 1
and B
= 1. Expected value E(X) of a single play, X = 1, is given
by
E(X) = ap - B:
E(1) = (1 X 0.48) - (1 X 0.44) =+0.04
With lots of plugs, strips and box cuts, the probability of
broken sequences was estimated as 0.23. Substituting this into
equation 7-2 yields:
P(h) = 0.38 - (0.23 + 0.10) = 0.05

Assuming E; = +0.51, E> = -0.34, E3 = -0.005 h = 0.05, d=
0.06 m= 0.88
E(X) = (0.51 X 0.05) + (-0.34 X 0.06) + (-0.005 X 0.89)

= 0.0255 - 0.0204 - 0.0044
= +0.00
The additional cuts have destroyed the profitability of this
game completely. This highlights the extreme importance of
finding shuffles with as few passes, riffles, and cuts, plugs,
strips or box cuts as possible.
PREDICTABILITY OF CASINO SHUFFLES
Do not believe the above expectation figures to be precise.

They are simply my best guesses based on probability theory.
Before predicting Aces in a real casino, you should test the
actual predictability of the shuffle using statistics. Ultimately,
as Epstein pointed out, mathematical expectation depends on
accuracy of prediction.
John May, a professional gambler, advised: “Many pro
players will try mentally to locate an Ace over a hundred times
before actually playing it ... Some teams will send in a
minimum bettor to ‘clock’ the dealer. This informant will
come away and communicate the information to his

confederates, who then analyze the data and send in a big
money team.”””
To collect the required data, the “clocker’ mentally
predicts an Ace one hundred times, betting the minimum. A
tally is kept of how many times the Ace lands on the predicted
box. This is compared to the number of successes predicted
by theory. If the success rate is higher than one in thirteen,
you have a better-than-guessing chance of predicting when an
Ace will be dealt.
In the six-deck, two-riffle R&R shuffle analyzed earlier 13
correct predictions are expected for every hundred made.
However, the actual number may be less, or more, than
expected. If it is less, but the game is still marginally
profitable, then you have two options: find a more predictable
shuffle, or play knowing you have a lower expectation. If you
opt for the latter, make sure you reduce your bet size to a safe
level.
The Exposed Ace
Occasionally a dealer will accidentally expose an Ace.

This usually happens when the dealer anticipates a hit, then
realizes the card is not required to finish the round.’ A
common casino practice is to place the exposed card under the
shoe, ready for the first hand of the next round.'®!
If you are sitting at first base, and know for sure your first
card will be an Ace, you should make a very large bet! Using
conventional basic strategy to play out the hand, Equation 7-4
yields the “optimal” fraction fof bankroll to wager:
f=EN 7-4
Where,
J= fraction of bankroll
E(X) = expectation on the hand
v = variance on the hand

Calculating the Variance
The variance’ for the outcome of this hand is greater than

a normal blackjack hand (average variance ~1.2)!% because
of the higher likelihood of a 3 to 2 payoff for a natural. For
expectation = +0.51, variance is calculated as follows:
TABLE 7-4. VARIANCE FOR THE EXPOSED ACE
eats Payoff eee Square Prob. Product
Win 1.50 0.99 0.98 0.48 0.47

Loss -1.00 -1.51 2.28 0.44 1.00
Push 0.00 -0.51 0.26 0.08 0.03
Totals ‘1.00
Thus,
f=+0.51/1.50=0.34
In other words, if you know an Ace will be your first card,
you should risk 34 percent of your bankroll. However, do this
only if you are certain you will get the Ace.
Unless an Ace is exposed as above, unpredictability caused
by card riffling, broken sequences and false key cards means
you cannot be absolutely certain where it will appear. It could
go to other players, or worse, the dealer. In such cases, bet
size should be much smaller.”
* See Chapter 8 of this book for a detailed discussion of optimal bet size when other
players or the dealer can get the Ace.

THE COUNT OF CIAMPINO
‘[s exiled King of Italy, Umberto II, waved his last

goodbye to a crowd of loyal supporters at Rome’s
Ciampino airport. As he boarded the plane, he
whispered to his financial attaché: “fa i conti” (do the
accounts). An appointments secretary, standing behind him,
thought he said: “make them counts.” Thus, two hundred new
Italian noblemen were created that day, all of them named
after the airport from which the King left Italy.
In 1973 Alessandro de Luca, Count of Ciampino, watched
in horror as his entire portfolio of stocks turned sour. He was
ruined. No longer able to afford his elegant rooms at
Monaco’s Hotel de Paris, he turned in desperation to the card
tables of Monte-Carlo, where, to his surprise, he discovered an
easy way to beat the blackjack game. Nearly every day for the
next 25 years, he exploited the same sweet secret.
At the Grand Casino, where croupiers spend 25 to 30 years
perfecting their skills, Alessandro knew every dealer by name
and, of course, they all knew him, but he had a special
fondness for the apprentice, an earnest, bespectacled young
man named Henri Renaud.
Alessandro would reserve seat No. 1 at Henri’s table when
it opened at 4 p.m. He had noticed Henri would slide a card
forward, out of the lip of the shoe, and then retract it when he
realized another card was not needed to complete the round.
Henri would place the card drawn in error underneath the
shoe, ready for the first hand of the next round. If it was an
Ace, Alessandro would bet 10,000 French francs, the
maximum his pocketbook allowed.
On the following round, with 100% certainty, Alessandro
would receive the accidentally exposed Ace, and was often
thrilled to see a Ten appear as his second card. Although he
had no idea how big his percentage was, his meticulously kept
accounts told him it was approximately half of everything he
bet.

For Alessandro this single bet, of which he made around
two hundred per year, constituted his entire day’s labor. Win,
or lose, he would gather up his chips, stroll through the
passage souterrain linking the casino to the hotel, and put on a
fresh shirt before dinner. Occasionally, if he had had a
winning day, he would allow himself a quiet smile, and toast
Henri with a bottle of Veuve Clicquot 1959.
119
Mathematical Expectation
How MUCH TO BET
EXPECTED RETURN
Expected return is the mean number of bets 4 you expect to

win over a given number of hands, based on the mathematical
expectation of the game:
j= mathematical expectation X number of bets (n) 8-1
For example, for one bet, n = 1: «= +0.04 X 1 bet = +0.04

bets.
For a series of n bets, multiply by n.
For example, for four bets, n = 4: uw = +0.04 X 4 bets =
+0.16 bets.
Frequency of Betting Opportunities
Blackjack expert Stanford Wong conducted a study of six-

deck shoe games to determine the average speed of a game.
On average, one player and the dealer take 12 seconds to play
out a single hand. Each additional player adds another 7
seconds to a single round. The average six-deck shuffle takes
100 seconds.'** Wong concluded that one player at a six-
decker gets an average of 248 rounds per hour, two players get
120 How Much to Bet

158, three players get 116, four get 91, five get 76, six get 64
and seven players get 56.'”°
In a six-deck game with 66% penetration and seven players
(excluding the dealer) you see an average of ten rounds per
shoe. The number of betting opportunities possible per shoe is
equal to the number of rounds per shoe minus one = 9. Fewer
players mean more rounds per shoe and more betting
opportunities. The random probability of a betting
opportunity depends on the number of riffles and whether the
sequence falls in the playable part of the shoe. With seven
players and the dealer, the minimum number of cards possible
for one round is 16. In a game where four of the six decks are
dealt, 208 cards are used before the cut card is reached.
However, when the cut card comes out in the middle of play
the round is completed. Thus, on average, another ((8 X 2.7) -
1) /2 = 10 cards will be seen. One card is deducted because,
if the cut card is poised to come out, the next hand is not
dealt.'°° Thus, a realistic estimate of the “upper” number of
cards possible for ten rounds is 218. This gives 218 - 16 =
202 cards dealt out as “playable” and 201 possible points
where the end of a round can occur. Actual betting
opportunities occur when the end of a round coincides with
the appearance of a key card but not the Ace that follows it.
The probability of an actual betting opportunity goes up if you
hold in memory more than one sequence per shoe. Holding
four sequences per shoe, the probability of a betting
opportunity in a two-riffle shuffle (with average three-card
separations of the type A|B|B|B|A) is given by 16 /201 = 0.08.
Table 8-1 below shows the average number of bets 7 you will
make per hour in a game with a two-riffle shuffle, assuming:
e A six-deck game with 66% of cards dealt before a

reshuffle.
e An average of 2.7 cards per hand.
e Number of players remains constant.
e Holding four sequences in your head from shoe to
shoe.
How Much to Bet 121

TABLE 8-1. NUMBER OF BETS PER HOUR
A Players at table (including the dealer) 8

B Cards per hand DS
C Cards per round (including the dealer)(4 XB) 21
D_ Cards dealt per shoe 218
E Rounds per shoe (D/ C) 10
F Shoes per hour 6
G_ Rounds per hour (E X F) 60
H_ Possible bets per shoe (£ — 1) 9
I Possible bets per hour (F X #) 54
J ACTUAL BETS PER HOUR (/ X 0.08) 4
Expected Return per Hour
n=4, p= 0.48, q = 0.44
u=+0.04 X 4 bets per hour = 0.16 ofa bet
RISK
The number of bets we will actually win or lose over a

series of wagers is more realistically predicted by the standard
deviation o, a measure of variability from the expected
number. For x trials of blackjack, standard deviation is given
bye”
o=1.1Vn 8-2
Where, m = number of bets
For n = 1, standard deviation per unit bet, o= ivn=1,3
Standard deviation per hour
For n = 4 bets, hourly standard deviation s, is:

S= 1,1N 4=999
bets
122 How Much to Bet

Dr. John L. Kelly Jr.'°* (1923-1965) proposed betting a
fraction of bankroll (total amount available for investment)
that maximizes bankroll growth rate and (in theory at least)
reduces risk of ruin to zero.
Kelly advised betting a decimal fraction k = p - q of
bankroll equal to the expectation on the next bet.'° The
player resizes the bet according to the current bankroll as
often as practically possible, and all winnings are reinvested.
Equation 8-3''° was used to calculate r the exponential rate
of growth of a bankroll for various fractions of k:
p= (k-1 / 200 Ly) 8-3
Where,
k = fraction of Kelly Bet
j= expected return per unit bet = +0.04
y = variance per unit bet = o°= 1.21
For example, for k= | (full Kelly)
r=(1-1°/2) X (0.047 / 1.21) = 0.000661157
TABLE 8-2. RELATIVE BANKROLL GROWTH RATE
fraction” k RGR
0.040 l I
0.036 0.9 0.98
0.032 0.8 0.94
0.028 0.7 0.88
0.024 0.6 0.80
0.020 0.5 0.71
0.016 0.4 0.60
0.012 0.3 0.48
0.008 0.2 0.34
0.004 0.1 0.18
* Fraction of bankroll wagered on a single bet.
123
How Much to Bet
Equation 8-4''' was used to compute the probability P(d)
of the bankroll doubling before halving:
Pia ea o 8-4
Where,
a=0.5
TABLE 8-3. DOUBLING BEFORE HALVING
fraction k Prob.
0.040 1 0.67
0.036 0.9 0.70
0.032 0.8 0.74
0.028 0.7 0.78
0.024 0.6 0.83
0.020 0.5 0.89
0.016 0.4 0.94
0.012 0.3 0.98
0.008 0.2 0.998
0.004 0.1 0.999
Equation 8-5''* was used to compute the probability P/h)

of ever losing half of the bankroll:
HO Ge 8-5
Where,
a
124 How Much to Bet

TABLE 8-4. EVER LOSING HALF
fraction k Prob.
0.040 1 0.5
0.036 0.9 0.43
0.032 0.8 0.35
0.028 0.7 0.28
0.024 0.6 0.20
0.020 0.5 0.13
0.016 0.4 0.06
0.012 0.3 0.02
0.008 0.2 0.002
0.004 0.1 0
Richard Epstein, in “An Optimal Gambling System For

Favorable Games,” (1964)''’ advised betting a fraction of
bankroll equal to half the expectation on the next bet (“4%
Kelly”).
Thorp commented: “My gambling and _ investment
experience, as well as reports from numerous blackjack
players and teams, suggests that most people strongly prefer
the increased safety and psychological comfort of ‘half Kelly’
(or some nearby value), in exchange for giving up 4 of their
growth rate.”''* For the professional Ace tracker playing at
very high stakes, hard experience suggests “’4 Kelly” (fraction
of Kelly Bet = 0.3) is most appropriate:
aD
Probability Bsaa
o &
8
o &6 3
ee—
i
| |
| |||
|
0.10
P(h) = 0/02 | | | |
0.00 | lead y
0.1 0.2 0.3 04 05 06 0.7 08 09 10
Optimal Fraction of Kelly Bet
FIGURE 8-1. OPTIMAL FRACTIONS OF A KELLY BET
How Much to Bet 125

Fig. 8-1 shows the bankroll will grow at 0.48 of the rate of
betting “full Kelly,” but the probability of the bankroll
doubling before halving is 0.98, while the chance of ever
losing half of it is only 0.02.
OPTIMAL BET SIZE
Equation 8-6''° was used to calculate x, the number of units

of bank required for any given fraction of a Kelly Bet:
x= s?/ ku 5-6
Where,
s = standard deviation per hour in bets
j4= expected return per hour in bets
For the “4 Kelly” bettor:
x= 2.27/03X0.16 =100 units
Optimal bet size
For E(X) = +0.04:

e Bankroll = 100 units
e +0.04 X 0.30 =+0.012
e¢ Optimal Bet Size ~0.01 X 100 units ~1 unit
Therefore, divide your bankroll by one hundred to get one

betting unit. For example:
Bankroll = US$40,000
126 How Much to Bet

Betting unit = US$40,000 = US$400
100
Betting units can be in any currency. One unit = US$40
0
or one unit = UK£400. Betting units can be of any value.
One
unit = US$40 or one unit = US$400.
Average Hands to Double or Halve
Equation 8-7''° yields the average number of hands N

before you either double the bankroll or lose half of it:
N= _1_ In(b*/a*") 8-7

r
Where,
r = exponential bankroll growth rate*
q = probability of doubling before halving
b=2
a=05
N= _1_ 1n(2°?8/0.5°"*') = 2051 hands

r
“DO NOT FORGET”
[:was after midnight when we left the casino, absolutely

giddy with excitement. Another big win!
Predicting Aces in Casino Baden-Baden’s American
Salon had turned out to be both easy and profitable. It was a
simple two-riffle R&R shuffle, with no “heat.” We’d won
nearly $20,000. It was time to leave town and bank the loot.
The next morning Dan and I went to the bahnhof to catch
the train to Ziirich. As it happened, every seat onboard was
* For this example, r= 0.000324433, k = 0.3
How Much to Bet 127

booked that day, but Dan, ever the optimist, had a clever, if
rather pricey, solution: “We’ll travel in the dining car.”
So, we found ourselves speeding to Switzerland, eating
gourmet food, sipping wine, reliving the exciting coup we'd
pulled off in Germany’s oldest and grandest gambling hall.
On the floor, under the table, sat Dan’s battered old black
attaché case. Inside was $60,000 in crisp hundred-dollar
bills—our $20,000 winnings plus the original $40,000 “joint
bankroll.”
We pulled into Ziirich about noon, got a tram to
ParadePlatz, then walked the few streets to Furst & Fuchs
Bank, where we found the gray-suited and bespectacled
manager, Herr Gruber, waiting for us in the marble lobby.
“Gentlemen,” he said, gesturing towards an oak-paneled door.
Sinking into Gruber’s plush office chairs, we relaxed for
the first time on the trip. Dan placed the case on the antique
desk in front of him, twirled it round to face Gruber, pressed
back the clasps and flicked open the lid. “We'd like to deposit
this in our numbered account,” he said. Gruber stared for a
moment or two, and then raised his eyebrows.
Dan jumped up and looked inside. His face froze.
Nothing!
“Oh my God, I’ve left the money in the hotel room!”
“Phone them up!” I almost shouted, but Dan shook his
head
“No.”
“What then?”
While Gruber sat open-mouthed, we figured the odds, from
“best” to “worst”:
A. The money was still in the hotel room.
B. A chambermaid had found it.
C. The next person to occupy the room had found it.
Ever the pessimist, my bet was “C.”
“We'll have to go back,” cried Dan. “Right now!”
Several hours later, after swallowing a tale about a
forgotten passport, the kindly clerk at our Baden-Baden hotel
turned the key in the door of Dan’s old room.
128 How Much to Bet

My nerves were in shreds as Dan groped gingerly into the
hole in the sink pedestal where he’d hidden the banknotes.
Eventually, he pulled out a black leather shoulder holster.
Inside, in three neatly packed wads, was $60,000. Attached to
the holster was a luggage ticket on which he’d carefully
printed in big red letters the words: “DO NOT FORGET.”
How Much to Bet 129

APPENDICES
APPENDIX A: SHUFFLING TERMINOLOGY
Discards The cards placed in the tray at the end of a

round.
Segment Any identified portion of the discards.
Slug A segment with extra high or low cards.
Stack Groups of shuffled cards placed one on top of
the other.
Pile Any portion of a stack.
Break To divide a stack or pile.
Grab To pick up a pile of cards with one hand.
Riffle To interleave two grabs.
Strip Shuffle The cards from the top of a grab are dealt on to
the table, reversing their order.
Box Cut The dealer makes a lopsided cut by hand.
Player Cut The player inserts the cut card into the final
stack.
Cutoffs The cards that end up behind the cut card after
the dealer’s cut.
Topped The cutoffs are placed on top of the discards.
Plugged The cutoffs are inserted into the middle of the
discards.
Bottomed The cutoffs are placed under the discards.
Riffle & Restack A shuffle where grabs are made on/y from the
original piles.
Stepladder A shuffle where grabs are made from the final
stack as well as the original piles.
Combo Shuffle A combination of two types of shuffles.
One-Pass The shuffle is over when the original piles are
gone and the final stack is complete.
Two-Pass The final stack is rebroken for a second shuffle.
Criss-Cross Piles are married using alternating grabs in an
“X” pattern.
130 Appendices
APPENDIX B: CARD MOVEMENTS IN PERFECT SHUFFLES
One perfect
Two perfect
jie 24 2 1S 28 AT
SG 29.) 42) 4 17 30 243
Se ts) ot 44 “6G 19 32. 45
+ 20°33 46 8 21 0934 47
S092 35 468 i)es) wWlon aS‘©
eS
No
De a+
ww on nN
Ww No
Two modified perfect
Aes? S14 1 4 28) 15

ae, 16 38 48, 80, 17 4
jameey 118. 3° 45 326 H9 6
Aeneas 20s T4734: 2h 28
ageess 27, 9-49" 36 23. 10
50 37 24 SiP e251 12
NnNO Ww\© N lon ee
we
Appendices we
Three perfect
Louesseel 4 Aue l, Kae eed

Te 345 Ago oe Oral 22
3°35 16. 48° 29 10 a
4 36 17 49°30 11 43 4
Sue 371s B50 kerries ees
Ge PE ICR OR ibs eos
ji 39" 20. 4.52
Three modified perfect
2940. 7 27 AT 1k 34,
DIP PA, 8.928) ASS isso: he
2242) 9 329 TA Ge6 a3
23g AS. 10. $30 “SOR igs es
DAN AA 11 73 ih eS) 1838 es
25° 45° 42 32°59) 1G 30" 6
132 Appendices
APPENDIX C: MEAN OF D;
MEAN OF D; (ONE-RIFFLE SHUFFLE)
D; Frequency Differences FXD DX (F XD)

0 271 -1 -271 271
1 1029 0 0 0
2 176 1 176 176
3 21 2 42 84
4 3 3} 9 27
1500 -44 558
MEAN OF D; (TWO-RIFFLE SHUFFLE)
D; Frequency Differences FXD DX

(F XD)
0 74 -3 -222 666
1 184 -2 -368 736
2 253 -1 -253 253
3} 550 0 0 0
4 211 1 Dall DIN
5 M5, 2 190 380
6 47 3 141 423
7 iL) 4 68 272
& 4 5 20 100
9 4 6 24 144
10 0 7 0 0
ipl ] 8 8 64
n= 1440 A= -181 B= 3249
Appendices 133
MEAN OF D, (THREE-RIFFLE SHUFFLE)
D; Frequency Differences FXD DX

(F XD)
0 45 -7 -315 2205
| 110 -660 3960
2 173 -865 4325
3 361 -1444 5776
4 278 2502
5) 439 1756
6 638
7 830
8 507
9 326
10 241
11 176
12 110
HS 59
14 38
US 31 fF
Nn
NHN
ery
CO
WN
—
16 13 DLT
17 5
18 9
LS) 4
20 2 Ssa&
oe xe
eS
gS
21 2
BY 1
23 0
24
25) 0
26 0
27 0
28 0
29 1 toi) 484
n = 4400 A =-1208 B= 41624
134 Appendices
APPENDIX D: x TEST
OBSERVED
Post-Shuffle Pre-Shuffle Position
Position 1-52 53-104 105-156 157-208 209-260 261-312
1-52 64 10 20 44 54 16
53-104 38 30 56 40 20
105-156 30 42 37 22 43
157-208 313) 30 29 14 49
209-260 13 36 29 26 33
261-312 30 50 13 52 47
EXPECTED 34.67
Diff?/Exp. 24.82 6.21 De 10.78 10.05
0.32 0.63 13.13 0.82 6.21
0.63 Ls) 0.16 4.63 2.00
0.08 0.63 0.93 12.32 S293
13.54 0.05 0.93 PEMA 0.08
0.63 6.78 13.54 8.67 4.39
DOSa/i
25
0.0000
Appendices 135
APPENDIX E: KENDALL’S CONCORDANCE COEFFICIENT
Data table for 2 deck averages

Iteration A B Cc D je) F H I K L
1 94.92 155.85 197.12 172.66 118.69 221.12 141.92 240.92 222.42 119.46
2 102.00 142.62 171.54 190.56 134.62 228.27 120.62 221.46 209.00 161.54
3 94.73 152.92 188.65 175.49 122.08 231-31 84.35 247.23 231.12 149.58
4 95.00 165.96 186.73 162.95 129.42 225.92 99.65 238.15 217.50 145.04
Ranks
Iteration I J K
il Zl 11
11 1 10
12 1 10
We = il 10
47 4 41
20 52026
2 = 22 el
441 484 225
136 Appendices
APPENDIX F: KRUSKAL-WALLIS TEST
ud A B Cc DE) ee? GP mo See

YR= 31 Ol) ase GSS O76 AT kT SO 16084
YR2= 961 10201 17689 15876 3364 28561 5776 2209 32761 100 25600 7056
N=48 4 4 4 ey eee oe Be ee eae ee
DR/n = 240 2550 4422 3969 841 7140 1444 552 8190 25 6400 1764
0.0051 X 37538 -147
rou 44.52
zy
df= l1
p<.01
Appendices we)
138 Appendices
INDEX
arithmetic mean, 46
A Armani, 77
Ace Asymmetric
location, 15, 16/17, 23. back designs, 103
2a eos 292 Ihe 1d, 19, face designs, 102
87 white space, 102
prediction, 15, 16, 17, Atlantic City, 28
ere2 2 2322930) average cards per hand,
31, 43, 77, 89, 104, it
114 average hands to double or
halve, 127
sequencing, 17
slug, 80
B
Tien Gosia? eZ,
73,101 Baccarat for the Clueless,
value as first card, 28, 26
29, 39, 108, 110 Baden-Baden, 128
Aces bankroll growth rate, 123,
average distribution of, 125
80 bankroll units, 116, 126
consecutive, 82 basic strategy, 96, 99, 100,
exposed, 116 TT 3G
probability of, 81, 105 Baxter, Thomas, 44
random, 97, 99 Bayer, David, 90, 91
Aces & Faces Blackjack, Beat the Dealer, 30, 34,
M5) 35, 108
ace-ten.com, 27 Bee® cards, 45, 66, 102,
advantage, 17, 18, 19, 21, 103, 104
99, 104, 108, 112 Bell Laboratories, 40
Aladdin Casino, 23 Best Blackjack, 25
Aldous, David, 91 Bets
amateur shuffle, 34 fractional, 123
An Optimal Gambling frequency of, 120
System For Favorable optimal, 116, 126
Games, 125 resizing, 123
Angeli, Michael, 25
Index 139
sizing, 116 Card Shuffling
betting unit, 126 Shenanigans, 28
big player, 18, 24 cardcounter.com, 29
binary digits, 41 Casino Baden-Baden, 127
bjmath.com, 31 Casino Card Shuffles
BJRnet.com, 16 How Random are
Blackjack Forum, 22, 23 They?, 19
Blackjack Forum Shuffle Casino de Genting, 28
Tracking Series, 25, 29 Casino Holiday, 108
Blackjack Reality, 22, 26 Casino Verité Shuffle, 20,
Blackjack Shuffle 21, G0, 61, 63, 70
Tracker’s Cookbook, Chi-squared test, 20, 69
Deja 22629 Chumash Casino Resort,
boxing, 21, 46, 55, 62, 115 29
Braun, Julian, 35, 109 Cigar Aficionado, 24
Bringing Down the House, Clark, Jeffrey, 19
28 clocking the dealer, 22,
broken sequences, 20, 21, 115
46, 47, 49, 51, 52, 54, Clump Reading and
OO; Ol sit his, Shuffle Tracking, 27
sg) coefficient of
Bujold, Laurent, 64, 65, determination, 76
104 Columbia University, 90
buzzards, 19 combinations, 98
computer simulations, 20,
C 70
confidence intervals, 46
Caesar’s Palace Casino, 25
California State count tracking, 24
University, 23, 33 counting the breaks, 55
card counting, 15, 23, 26, Curtis, Anthony, 39
2830535 cutcardAss "81202. 105:
Card Counting
for the 121
Casino Executive, 23
cutofts, 54.55/56, 662.
card patterns, 94 Perec Site
Card pros count more than
money, 24 D
Card Sequencing, 27 Dalton, Michael, 15
Danese, Roseann, 24
140 Index
De Luca, Alessandro, 118 G
Diaconis, Persi, 90, 91
Gambling Times Guide to
dilution shuffle, 67
Blackjack, The, 23
diseards.21 27, 28, 55,
game speed, 120
2656). 625 12.73,.79;
geometric distribution, 37
80, 83, 92, 96, 100, 104
Get the Edge at Blackjack,
doubling after splits, 110
22, 26
Downing, Douglas, 19
Gilbert, Edgar, 19, 40
Golomb, Solomon, 19, 32
E
grab, 54, 55, 81, 83, 84,
Emanuel, David, 40 85, 86, 87
Encyclopedia of Casino Grand Casino de Monte-
Twenty-One, 16 Carlo, 118
Epstein, Richard, 19, 35, Griffin, Peter, 33, 108
39098, 3B9542789,90; Gwynn, John, 23
012-942-9771 Is, 125
Esquire, 25 H
expectation, 89, 109, 110,
Hacking Las Vegas, 28
WiehS hlS 1202123 Hannum, Robert, 19, 39,
expectation formula, 17,
40, 66
2 AAS hh He had the upper hand,
expected return, 120 now all bets are off, 29
expected return per hour, Nittate, 212 ene
122 lig
eyeballing, 62, 87, 105 Hoffman, Professor Louis,
43
F Hotel de Paris, 118
false key card, 29, 80, 100, How to Develop a Perfect
Lote Ol Tir tS eh? Memory, 24
Feller, William, 34 How to Make $100,000 a
Finley, Arthur, 44 Year Gambling for a
Fleecing Las Vegas, 25 Living, 26
Forgotten Statistics, 19 How to win at blackjack,
Forte, Steve, 22, 103 pM
hypergeometric
distribution, 81, 98, 101
Index
141
I lay-and-pay, 29
Learned Pig Magic eZine,
imperfect shuffle, 34, 37, 44
41,42 Learning Simulation
in shuffle, 33 Techniques on a
infinite deck, 38 Microcomputer Playing
informational entropy, 41 Blackjack and Other
intact sequences, 20, 54, Monte Carlo Games, 34
60, 61, 64 Learning Statistics
Through Playing Cards,
J
19
Jacobson, Eliot, 29 Lewis, Angelo, 43
Jordan, Charles, 43, 44 Lewis, Kevin, 28
Life, 30
K line of best fit, 74, 75
loading up, 86
Kamelhair, Abigail, 14
location rule, 80
Kelly, John, 123
Loren, Sophia, 87
Kendall’s Coefficient of
Concordance, 20, 71
M
key card #20521" ie 72)
80, 92, 93, 96, 99, 100, Macaluso, Pat, 34
1027 105; L06ML ML 1 Magical Bulletin, 43
alignment, 102 Malaysia, 28
memorizing, 24, 25, 27, Malmuth, Mason, 26
28, 29, 86,92, 965.97, Man vs. Computer: Does
99100. 102,+921 Casino Blackjack Differ
sequencing, 17, 21; 22) From Computer-
POR Simulated Blackjack?,
kitchen table analysis, 72 23
Knapp, Thomas, 19 Marko, 44
Knuth, Donald, 35 Massachusetts Institute of
Konik, Michael, 24 Technology, 28
Kruskal-Wallis Test, 71 Mathematics of Gambling,
The, 31
L May, John, 22, 26, 115
Las Vegas, 17925; 30937,
McGarvey, Rob, 27
Mezrich, Ben, 28
39, 66
Las Vegas Advisor, 39 Mississippi, 28
142 Index
MIT blackjack team, 17, Observer Sport Monthly,
28 2128
modified perfect shuffle, One Step Ahead, 24
33, 80 O'Neil, Paul, 30
Monaco, 118 one-pass shuffle, 21, 55,
Monte-Carlo, 118 83, 85
Morse, David, 22, 26 out shuffle, 32
Murray State University,
40 P
patterns of prediction, 94
N
perfect shuffle, 32, 33, 36,
natural, 117 42
negative decks, 99 permutations, 82
neutral decks, 110 Permutations by Cutting
New Mexico State and Shuffling, 32
University, 30 Peterson, Ivars, 28
Newman, Sandra, 28 phantom Aces, 111
Nice, 64 pick-and-pay, 29
Nikolopoulos, George, 87, player cut, 59, 60, 64, 79,
88 86, 93
Noir, Jacques, 108 playing the turn, 103
Nonparametric Statistics plugging, 21, 46, 54, 55,
for the Behavioral OlFG2; 8% 115
Sciences, 19 pointer cards, 102
non-random shuffling, 19, Port de la Santé, 64
23250131032, 34,35, power series shuffle, 67
40, 42, 68, 69, 72, 89 prediction bands, 77
Non-Random Shuffling prediction intervals, 77
Strategies in Blackjack, prediction rule, 96
40 Princeton University, 34
Nonrandom Shuffling with
Applications to the R
Game of Faro, 31, 35 Reader of the pack, 28
normal curve, 46
reading the edges, 104
regression analysis, 74
O
Reid, Richard, 31
O’Brien, Dominic, 24 Renaud, Henri, 118
Index 143
riffle-and-restack shuffle, stack, 20, 54, 87
27,30; 54,.55,,00,61; staggered riffle shuffle, 67
Ol, Sole! Lane standard deviation, 46, 122
(27 standard deviation per
rising sequences, 44, 54 hour, 122
risk, 1225123 Stanford University, 90
risk of ruin, 123 Steele, W. F. Rufus, 44
roulette, 30 steering, 24, 29, 99, 100
Ruffled by the Shuffle, 23 stepladder shuffle, 29, 67,
83, 85
S stripping, 21, 46, 55, 61,
o2za115
Sands Casino, 77
stutter shuffle, 67
scatter plot, 68, 72, 73
Sutrick, Kenneth, 40
Science News, 28
Switzerland, 128
Scoblete, Frank, 25
sequence tracking, 24, 25
T
Shannon, Claude, 30, 41
shuffle map, 21, 73, 84, 87 Tamburin, Henry, 27
shuffle tracking, 15, 23, 28 Tangier, 65
shuffle weaknesses, 20, 70, target card, 24, 79, 92, 93,
T2138), Tae 1 Oe 1 Oe LO4 95, 100
Siegel, Sidney, 19 memorizing, 97, 99
signature, 43, 66, 113 Ten
signature card, 26, 92 as second card, 110
Sinatra, Frank, 77 value as first card, 108,
skilled predictor, 112 109
Sklansky, David, 26 Test for randomness, 69
slug, 77, 80 Thayer, Floyd, 43
Snyder, Arnold, 21, 22, 23, Theory of Blackjack, The,
25,2931 280, 87.99, 33
108; 113 Theory of Gambling and
Snyder’s rule of thumb, Statistical Logic, The,
Ms eil4 89
soft doubling, 110 Thirty Card Mysteries, 43
squared error about the Thorp, Edward, 19, 23, 30,
average, 75 31, 34, 35, 37, 45, 46,
squared error about the LO8ei25
INtaen gS
144 Index
Trailing the Dovetail Vv
Shuffle to Its Lair, 43,
variance, 116
90
Veuve Clicquot, 119
TropWorld Casino, 28
Villefranche-sur-Mer, 64
two-pass combo shuffle,
21, 67, 68, 69, 85, 104
Ww
Tyche, 88
Walden, William, 31, 35
U Wattenberger, Norman, 20,
21
Umberto I, 118
Williams, Charles, 43
uncertainty, 41
Windsor Star, 24
University of California,
Wired, 28
DAS SaNh
Wong, Stanford, 120
University of Denver, 39
University of Southern
xX
California, 32
Unlocking the Shuffle LL Statistics. 15,76
Trade Secrets of the
Shuffle Trackers, 25 Z
unskilled predictor, 112 Zender, Bill, 23
zone shuffle, 55
Ziti ele ae 8
Index
145
REFERENCES
' HANNUM, ROBERT C., “Casino Card Shuffles: How Random

are They?” in Finding the Edge: Mathematical Analysis of
Casino Games, Reno, Nevada: Institute for the Study of
Gambling and Commercial Gaming, University of Nevada,
July, 2000, pp. 33-55.
* DOWNING, DOUGLAS, AND JEFFREY CLARK, Forgotten

Statistics: A Refresher Course With Applications to Economics
and Business. Hauppauge, New York: Barron’s Educational
Sericss ne. 1.996)
> KNAPP, THOMAS R., Learning Statistics Through Playing

Cards. Thousand Oaks, California: SAGE Publications, Inc.,
1996.
* SIEGEL, SIDNEY, Nonparametric Statistics for the Behavioral

Sciences. New York: McGraw-Hill Book Company, 1956.
> SNYDER, ARNOLD, The Blackjack Shuffle Tracker’s

Cookbook. Las Vegas, Nevada: Huntington Press, 2003.
2 , Jrade Secrets of the Shuffle Trackers Part III,”

Blackjack Forum, Vol. XV, #1, March, 1995, pp. 19-29.
” SNYDER, 2003, pp. 74-77.
: Morsk, DAVID H., Blackjack Reality: The David H. Morse

Method for Winning at Blackjack. Self published, 1998, pp.
79-86.
» MAY, JOHN, Get the Edge at Blackjack. Chicago: Illinois:

Bonus Books, Inc., 2000, pp. 47-59.
2 ROBERTS, STANLEY, et al., The Gambling Times Guide to

Blackjack. Secaucus, New Jersey: Lyle Stuart, September,
146 References
1984, Chapter 3, “The Principles of the Game and Why It Can
Be Beaten,” p. 30.
'' CATLIN, DON, “Remembering John Gwynn,” The Rolling

Good Times Online, www.scoblete.com, July 1, 2001.
'* GWYNN, JOHN M. JR. AND ARNOLD SNYDER, “Man vs.

Computer: Does Casino Blackjack Differ From Computer-
Simulated Blackjack?,” Blackjack Forum, Vol. VIII #1,
March, 1988, p. 6.
'° SNYDER, ARNOLD, “Ruffled by the Shuffle,” Blackjack

Forum, Vol. X, #1, March , 1990, p. 5.
'4 ZENDER, BILL, Card Counting for the Casino Executive.

Las Vegas, Nevada: Self published, 1990, p. 93.
'° O’BRIEN, DOMINIC, How to Develop a Perfect Memory.

London: England: Pavilion Books Limited, 1993, p. 144.
'© KONIK, MICHAEL, “One Step Ahead: Sophisticated

Gamblers Use Legal Techniques to Gain Small Advantages at
Casino Games,” www.cigaraficionado.com, Archives,
Wednesday, March 1, 1995.
'’ DANESE, ROSEANN, “Card pros count more than money,”

Windsor, Ontario: The Windsor Star, Friday, July 21, 1995.
'8 SNYDER, 1995, p. 19.
'9 SCOBLETE, FRANK, Best Blackjack. Chicago, Illinois: Bonus

Books, Inc., 1996, p. 175.
20 ANGELI, MICHAEL, “Fleecing Las Vegas; blackjack team,”

Esquire, May 1997, No. 5, Vol. 127, p. 63.
References 147
21 SKLANSKY, DAVID AND MASON MALMUTH, How To Make
$100,000 a Year Gambling for a Living. Henderson, Nevada:
Two Plus Two Publishing, 1° ed., 1997, pp. 55-S6.
2 MORSE, 1998, p. 1.
*3 MAY, JOHN, Baccarat for the Clueless. Secaucus, New

Jersey: Carol Publishing Group, 1998, p. 68.
- , Get the Edge at Blackjack. Chicago: Illinois:

Bonus Books, Inc., 2000, p. 47.
25 TAMBURIN, HENRY, “New Blackjack Winning

Techniques—Card Sequencing,” www.casinoheadliners.com,
Monday, October 15, 2001.
26 MCGARVEY, ROB, Aces & Faces Blackjack. Self published,

May, 2001, p. 83.
*7 “ACE”, “Clump Reading and Shuffle Tracking,” www.ace-

ten.com, 2001.
*8 NEWMAN, SANDRA, “How to win at blackjack,” Observer

Sport Monthly, http://observer.guardian.co.uk, Sunday, April
T2002;
zs , ‘Reader of the pack,” Observer Sport Monthly,

http://observer.guardian.co.uk, Sunday, April 7, 2002.
°° MEZRICH, BEN, Bringing Down the House: The Inside Story

of Six MIT Students Who Took Vegas for Millions. New Y
ork,
New York: The Free Press, 2002, “How to Count Cards and
Beat Vegas,” p. 257.
“ , “Hacking Las Vegas,” Wired, Issue 10.09,

September, 2002.
148 References
oe PETERSON, IVARS, “Card Shuffling Shenanigans,” Science
News, Vol. 162, No. 20, Saturday, 16 November, 2002.
*° SNYDER, 2003, pp. 74-77.
** ETLING, LEAH, “He had the upper hand, now all bets are
off,” Santa Barbara News-Press, Sunday, November 9, 2003.
*° JACOBSON, ELIOT, “The Mayor’s Podium: Ace location

techniques,” www.cardcounter.com, May, 2004.
O’NEIL, PAUL, “The Professor Who Breaks the Bank,” Life,

Chicago, Illinois: Time, Inc., March 27, 1964, p. 84.
a CALDERBANK, ROBERT AND NEIL J. A. SLOANE,

“Obituary: Claude Shannon,” Nature, Vol. 410, No. 768,
2001.
°8 THORP, EDWARD O., Beat the Dealer: A Winning Strategy

for the Game of Twenty-One, New York: Blaisdell Publishing
Company, 1* ed., 1962.
*° O’NEIL, 1964, p. 91.
40 THORP, EDWARD O., “Nonrandom Shuffling with

Applications to the Game of Faro,” Journal of the American
Statistical Association, Vol. 68, No. 344, December, 1973, p.
844.
4! THORP, EDWARD O. AND WILLIAM E. WALDEN, “The

Solution of Games by Computer,” (unpublished manuscript),
1963.
” THORP, 1973, p. 464.
" , The Mathematics of Gambling. Secaucus, New

Jersey: Lyle Stuart, March, 1985.
References 149
44 GOLOMB, SOLOMON W., “Permutations by Cutting and
Shuffling”, SIAM Review, Vol. 3, No. 4, October, 1961, p.
293:
4S THOMAS, ROBERT MCG., JR., “Peter Griffin, Solver of

Blackjack, Dies at 61,” The New York Times, Obituaries,
Monday, November 2, 1998.
© GRIFFIN, PETER A., The Theory of Blackjack (Revised and

expanded). Las Vegas, Nevada: Gambler’s Book Club Press,
Oeed BOS ape 30.
“7 MACALUSO, PAT, Learning Simulation Techniques on a

Microcomputer Playing Blackjack and Other Monte Carlo
Games, Blue Ridge Summit, Pennsylvania: TAB BOOKS
Mesos Sp Os
“8 THORP, 1962, p. 171.
*° MACTUTOR HISTORY OF MATHEMATICS ARCHIVE, THE,

www.st-andrews.ac.uk, “Biographies—Feller, | William,”
School of Mathematics and Statistics, University of St.
Andrews, Fife, Scotland.
°° FELLER, WILLIAM K., An Introduction to Probability

Theory and Its Applications, Vol. 1, New York: John Wiley &
Sons Inc., 1957, p. 368.
>! THORP AND WALDEN, 1963.
°? RUCHMAN, PETER, “Not Fade Away: An Appreciation of

Julian Braun (1929-2000), Part II,” www.casinogaming.com,
Sunday, March 4, 2001.
150 References
- KNUTH, DONALD E., The Art of Computer Programming.
Reading, Massachusetts: Addison-Wesley Publishing
Company, Inc., 1969, pp. 124-151.
> THORP, 1973, p. 844.
a , 1973, p. 845.
°° EPSTEIN, RICHARD A., The Theory of Gambling and

Statistical Logic. San Diego, California: Academic Press,
L977, ps 164.
”” Ibid.
°8 GWYNN AND SNYDER, 1988, p. 7.
>? HANNUM, 2000, pp. 51-52.
6° EMANUEL, DAVID C. AND KENNETH H. SUTRICK, “Non-

random Shuffling Strategies in Blackjack,” Communications
in Statistics: Theory and Methods, Vol. 17, Issue 9, 1988, p.
2954.
°" [bid.
6? GILBERT, EDGAR N., “Theory of shuffling,” Technical
Memorandum, Bell Laboratories, Murray Hill, New Jersey,
1955.
© EPSTEIN, 1977, p. 167.
64 TREFETHAN, LLOYD N., AND LLOYD M. TREFETHAN, “How

many shuffles to randomize a deck of cards?” Proceedings of
the Royal Society of London, Series A, 1999.
6 EPSTEIN, 1977, pp. 167-168.
References 151
66 MAY, 2000,
p. 51.
6? FULVES, KARL, Charles Jordan’s Best Card Tricks.

Toronto, Ontario: General Publishing Company, Ltd., 1992,
pp. v—vi.
68 BAYER, DAVID, AND PERSI DIACONIS, “Trailing the

Dovetail Shuffle to its Lair,’ The Annals of Applied
Probability, Vol. 2, No. 2, May, 1992, p. 299.
© JORDAN, CHARLES T., “Trailing the Dovetail Shuffle to Its

Lair,” in Thirty Card Mysteries, reprinted Oakland, California:
Magic Limited—Lloyd E. Jones, 3" ed., 1974, p. 1.
7° MONTICUP, PETER, Magic Tricks Inc., Gordonsville,

Virginia: www.magictricks.com, Library, Biographies, 1996—
2003.
™ FULVES,
1992, p. 114.
” LEwIS, ANGELO, “Scraps from My Note Book, by Professor

Hoffman, No.l—Card Reading Extraordinary,” London,
England: Gamage’s The Magician Monthly, 8, 1912, p. 67.
” BAXTER, THOMAS, AND “MARKO”, “True Men of Mystery,”

The Learned Pig Magic eZine, Vol. 1, No. 4, Tuesday,
November 30, 1999.
™ THORP,
1973, p. 845.
SIEGEL,
1956, pp. 229-238.
”® Ibid.
77
, 2003,
p.97.
’8 EPSTEIN,
1977, p. 218.
152 References
79
, 1977,
p. 249.
o , “Card Prediction for A Shuffled Deck,” Hughes
Aircraft Co., El Segundo, California, Report TP-64-19-11,
OP-68, July, 1964, p. 1.
81
OTIS pal67:
*? BAYER AND DIACONIS, 1992, p. 294.
8° ALDOUS, DAVID, AND PERSI DIACONIS, “Shuffling Cards

and Stopping Times,” American Mathematical Monthly, Vol.
93, May, 1986, p. 345.
8* EPSTEIN, 1977, pp. 167-171.
= , 1977, p. 179.
8° ZENDER, 1990, pp. 94-95.
87 SNYDER, 2003, p. 75.
88 KONIK, 1995.
8°Ibid.
°° THORP, 1962, p. 98.
°! Norr, JACQUES, Casino Holiday. Berkeley, California:

Oxford Street Press, 2" ed., 1970, p. 127.
°2 GRIFFIN, 1981, pp. 145-146.
°3 SNYDER, 2003, p. 76.
References 153
°4 BRAUN, JULIAN H., How to Play Winning Blackjack.
Chicago, Illinois: Data House Publishing Company, Inc.,
1980, pp. 82, 83.
°° GROSJEAN, JAMES, AND PREVIN MANKODI, “More on the

Ace in Hand ...,” Blackjack Forum, Vol. XXIII, #4,
December, 2003, pp. 51-64.
°° SNYDER, 2003, p. 76.
°7 EPSTEIN, 1977, p. 234.
°8 WILSON, ALLAN N., The Casino Gambler’s Guide. New

York: Harper & Row, Publishers, Inc., 1970, p. 289.
”” MAY, 2000, pp. 55-56.
'° GROSJEAN, JAMES, Beyond Counting. Oakland, California:

RGE Publishing, 2000, p. 93.
'°' MAY, 2000, p. 63.
'0? WONG, STANFORD, Professional Blackjack. New York,

New York: William Morrow and Company, Inc., 1981, p. 91.
ne , 1981, p. 93.
104
, Professional Blackjack. Las Vegas, Nevada: Pi
Yee Press 19945 pa245:
_ , 1994, p. 237.
°° SCHLESINGER, DONALD, Blackjack Attack: Playing the

Pros’ Way. Las Vegas, Nevada: RGE Publishing, Ltd., 3"
ed., 2004, p. 18.
gl , 2004, p. 16.
154 References
°° BERLEKAMP, ELWYN, R., “Shannon Lecture,” Institute of
Electrical and Electronics Engineers: — International
Symposium on Information Theory, San Antonio, Texas,
January 17—22, 1993.
' KELLY, JOHN L. JR., “A New Interpretation of Information

Rate,” Bell System Technical Journal, 35, No. 4 (July, 1956),
Page:
''° CHIN, WILLIAM AND Marc J. INGENOSO, “Risk Formulas

for Proportional Betting,” www.bjmath.com, March 13, 2004,
p. 4.
He 92004165,
ao Ibid.
''S EPSTEIN, RICHARD A., “An Optimal Gambling System For

Favorable Games,” Hughes Aircraft Co., Report TP-64-19-26,
OP-61, July, 1964, pp. 2—S.
''4 THORP, EDWARD O., “The Kelly Criterion in Blackjack,

Sports Betting, and the Stock Market,” paper presented at the
Tenth International Conference on Gambling and Risk Taking,
Montreal, June 1997; revised Friday, May 29, 1998, p. 32.
''S CHIN AND INGENOSO, 2004, p. 9.
116 , 2004, p. 5.
References 155
Gaining an edge in cards and life!
Many of the books, software and other products

mentioned in this book are available in our full online
catalog at:
www.BJRnet.com
The Blackjack Review Network is the _ premier

blackjack, poker, gambling, and card _ counting
information site for players who can think. We have
dozens of message boards, chat rooms, gambling
archives, free links, and a full product catalog for the
serious player. We are not affiliated with any casino
operation and are devoted to providing honest and
accurate information that players can use to beat (i.e.,
gain a theoretical advantage in) today's tough casino
games. While visiting our site be sure and check out our
“21 Frequently Asked Questions About Blackjack” — a
must read for new players.
The Blackjack Review Network is also the

home of Michael Dalton’s Encyclopedia of
Casino Twenty-One -—- _ an_ exhaustive
“member-only” online resource covering
everything anyone would ever want to know
about the game.

Spur of the Moment Publishing
PO BOX 541967
Merritt Island, FL 32954-1967 USA
orders@BJRnet.com |www.BJRnet.com
Casino Verite Blackjack, on 0> ¢
CVData & CVShuffle by QFIT
Casino Verite Blackjack (CVBJ) is the best profess
ional blackjack
game and simulation software on the market today. Contain
s over 100
rule and play variations and a database of over 890 rule
sets in over
500 real casinos. This amazing software also includes
maps to your
favorite casino, different table formats, statistic logs and color
graphs,
tournament play, counting drills, voice recognition, custom strategi
es,
unusual games and more. If the strategy that you use is not included
,
you can define your own in excruciating detail. Strategy, betting
and
Over/Under 13 strategies can all be user defined. Warnings and
records of strategy violations are provided. New variations of
Blackjack are supported, like Multi-action, Over/Under13, Red/Black,
Double Exposure, Royal Match, Bust Out, Super 7's and Seven &
ts
“Casino Verite Blackjack get's my pick for the "best” overall

blackjack software package on the market. Outstanding!”
— Michael Dalton / Blackjack Review Network.
CVCX is a Combination Simulator, Sim Library and Risk / Goal /

Bankroll / Optimal Bet calculator for advanced players. CVCX is a
completely new application providing the ability to create your own
customized, interactive Blackjack Attack Chapter Ten type tables.
If you want to perform advanced blackjack simulations you need

CVData. This software is the replacement product for CVSim.
CVSim users will recognize the output section. However, there are
zero lines of code in common between the CVSim and CVData BJ
engines. CVData's structure, goals and technology are very different.
CVShuffle is a set of additions to Casino Vérité Blackjack in support

of shuffle tracking practice and analysis. CVShuffle is a separate
product that requires a minimum CVBJ Version 3 release. Shuffle
tracking is an extension to card counting. You not only count cards as
they go by, but you also take note of the location of large numbers of
large or small cards in the discard tray to determine or control their
location in the next shoe. Not all shuffles are ‘trackable.’ CVShuffle
includes aides to determine the ‘trackability’ of shuffles as well as
practice aides and tools to help beginners understand the concepts.
For availability and the latest pricing on this

incredible software visit the Blackjack Review
Network online at www.BJRnet.com/shop/
The Hot Shoe
y\9 DAVAD Eyol’ BEAVA[emar: h¥4ce)4]
The Hot Shoe is the best documentary movie about blackjack
card counting ever produced. By David Layton. Executive
Producers Michael Berkaw and Bill Petrick. Music by Dylan
Keefe. The history, the mathematics, the politics and the
practical realities of blackjack are addressed in_ this
documentary, the most complete and accurate treatment of
card counting ever made. Interviews with blackjack experts,
casino managers, authors, detectives, a legal expert, and the
most legendary players in the world. Behind-the-scenes
footage of casino pits, surveillance rooms, members of the MIT
card counting team strategizing in and playing in casinos
during a busy Super Bowl weekend (shot with hidden
cameras). Demonstration of the high-tech methods that
casinos and detective agencies use to identify and track card
counters, including facial recognition technology.
A case study, wherein the

director of the movie learns
how to count cards and takes
the $5000 that was budgeted
for the soundtrack and uses it
as a stake to play blackjack in
> Las Vegas over ten days.
a Be
es
ST a m includes interviews and video
gt © eee of: Gordon Adams - Griffin
es. Ewe ng Investigations, lan Andersen -
tex & Author, Burning the Tables in
rie Las Vegas, Andy Anderson -
ra SEE Casino Visual Identification,
GaueoF 2I Andy Bloch - MIT team, John
by Davin LAYTON Chang - MIT team, Richard
inners §Chen - M/T team, Anthony
Curtis - Publisher of
Huntington Press, Mark Estes - Former card counter, Steve
Forte - Author, Read the Dealer, Peter Griffin - Author, The
Theory of Blackjack, Tommy Hyland - Card counter, Jim Kilby -
UNLV, Nelson Rose - Author, Blackjack and the Law, Max
Rubin - Author, Comp City, Howard Schwartz - Gambler's
Book Shop, Arnold Snyder - Author, Blackbelt in Blackjack,
Ralph Stricker - Card counter and author, Ed Thorp - Author,
Beat the Dealer, Olaf Vancura - Author, Knock-Out Blackjack, —
Stanford Wong - Author, Professional Blackjack, Bill Zender -
Author, Card Counting for the Casino Executive.
“The Hot Shoe DVD wins my editor’s choice for

the best blackjack documentary.”
— Michael Dalton / Blackjack Review Network
A MESSAGE FROM THE DIRECTOR: /'m very proud to

present "The Hot Shoe," a movie | believe to be the most
complete and accurate documentary about card counting ever
made. Most of the specials on television that deal with the
subject have been incomplete at best and sometimes totally
inaccurate. | struggled to make a movie that would deal with
the subject in all its complexities and yet still remain interesting
to people not already familiar with card counting. The
mathematics, the history, the politics and the practical realities
of blackjack all feature prominently in this movie. | talked to the
leading lights in the world of blackjack—professional players,
authors, casino personnel, detectives who identify players,
lawyers and academics. If you want to know the real story, if
you want to learn how to count cards, or if you just want to
watch a fun adventure | think you will enjoy "The Hot Shoe."
To order the Hot Shoe DVD by David Layton in the

USA send a $29.99 money order or bank draft (no
personal checks please) to:
Spur of the Moment Publishing

PO BOX 541967
Merritt Island, FL 32954-1967 USA
This product can also be ordered online using credit

card or PayPal at:
www.BJRnet.com/shop/
ATTENTION: Non-USA order requests can only be handled using our secure
online ordering system. The Hot Shoe DVD is compatible world-wide.
= a= = ne Le
Blackjack Blackjack Blackjack Blackjack Blackjack BlackJack Blackjack Blackjack ‘Blackjack

Blackjack Blackjack
EW Vow waa §
uae ik
&@ === 4& 4s
Dalton’s Blackjack Review Magazine officially ended

publication in 1998. Twenty-three issues were published over
seven years during
SPRING 1997 www .BJAnet.com PAT
VYOCUWL & SUE ? BS 55 IN CANADA) the biggest casino
gambling revival in
United States
history. Player
reports and casino
blackjack rules and
conditions were
the heart of each
issue but it also
WORLDWIDE BLACKJACK NETWORK attracted some of
RULES
Ovee
AND
BSO
TABLE
Casinos
CONDITIONS
LISTED
the best minds and
writers in the
Dow Goran's
industry. Each
SHUFFLE TRACKING quarterly issue
averaged about 50
Using Neural NETWORKS
pages and
Rirz, Ryv, axe
Tere Stace WrecoeT
me OMSCARD peat
woke mash messin
included
ty Berce Canccow is no-nonsense infor-
Greuntwe rie Sewerus mation on playing
sy David Morse and beating our
Now Bteca Car Yor War?
& Tae Sra ov Poacey Cuaarrae
favorite casino
ay Adscwaci Dalrow game.
Unfortunately, very
CASINO Best Bets, LetTrers, Proouct REVIEWS,
INTERNET SUMMARIES & few full collections
PLAYER REPORTS FROM AROUND THE WoRLD
remain. Visit
www.bjrnet.com/
shop/backissues.htm for details on how to order the last
remaining back issues and the few full collections that may still
exist. Better hurry, before they are all gone.
a, ee ———- = a
Blackjack Blackjack Blackjack Blackjack Blackjack (ier tias Blackjack Blackjack Blackjack Blackjack Blackjack
~Review Review Review Review Review ‘eview “Review - Review Review ~Review Review
ee CeCETEES
THE ART OF ADVANCED LOCATION
STRATEGIES FOR THE CASINO
GAME OF TWENTY-ONE!
"With his compelling Blackjack Ace Prediction, David McDowell, a

relative unknown in the blackjack world, bursts onto the scene in
impressive fashion. His scholarly approach to the subject ... is ajoy to
read. | have a feeling that the relatively small group of players already
aware of the techniques that McDowell describes are not going to be happy
that he has gone ‘mainstream’ with the information. This is a well-researched
treatise, and I'm happy to recommend it to serious students of the game.”
Don Schlesinger — author of Blackjack Attack: Playing the Pros’ Way
"If you are interested in how to beat casinos by analyzing non-random

shuffling, you will find David McDowell's Blackjack Ace Prediction an
interesting and enjoyable read."
Ed Thorp — author of the best selling book Beat the Dealer
"Many will consider McDowell’s book as the final chapter in advan-

tageous blackjack play — as the cat has finally been let out of the bag.”
Michael Dalton — author of The Encyclopedia of Casino Twenty-One
"David McDowell's Blackjack Ace Prediction is a winner. It will, no doubt,

prove to be the best work on ace tracking published to date.”
Steve Forte — author of the classic book on tells, Read the Dealer,
and the Gambling Protection Series videos.
"David McDowell provides an interesting example of how the game

of blackjack can still be exploited by the knowledgeable player. This
book is a must read for any player interested in assaulting the
casinos by locating and tracking those valuable aces."
Bill Zender — author of Card Counting for the Casino Executive
“David McDowell, the author, and Michael Dalton, the publisher, are to
be congratulated for this important addition to our knowledge of blackjack."
Lance Humble — senior author of The World's Greatest Blackjack Book
COVER ART BY ISBN 1-8?9?12-10-5
ABIGAIL KAMELHAIR 5 3,4 95>
FOREWORD BY
9"781879"712102
MICHAEL DALTON ISBN: 1-879712-10-5 $34.95 U.S «8Z9-
TH
TT

McDowell, David - Blackjack Ace Prediction. The Art of Advanced Location Strategies

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

McDowell, David - Blackjack Ace Prediction. The Art of Advanced Location Strategies

Uploaded by

Copyright:

Available Formats

- BLACKJACK

With a foreword by Michael Dalton

09 08 07 06 05 04 OSS]. Guat See) 1

Includes appendices, bibliographical references and index.

All rights reserved. No part of this publication may be

Visit our web site at www.BJRnet.com to order copies of this

Blackjack Review Network

Michael Dalton, Spur othe Moment Publishing, Merritt Island,

COVER ARTWORK CREDIT:

Copyright © Directional Publishing, Inc., “Ace of Spades” by Abigail T.

Edward Thorp, courtesy of New Mexico State University Library Archives,

William Earl Walden, courtesy of Washington State University Library

Edward Thorp at University of California, Irvine, from I Have a

Shoulder holster, courtesy of Warren of California, San Diego,

Although the author and publisher have exhaustively

Gambling infers risk. The author and publisher accept no

Tables and Figures icc.::-sscsaceesesranscese.

About the Cover: Art sasiizi.scssstevsaceechsaceutsvscescsvetcacweseereneeds 14

Foreword by Michael Dalton .............c.sscsccssscssessesccenscenees Lbs

Chapter] =A. ce Prediction i... s1ecc hess nae coe ee 17

Doing ShufflesAnalyses- Seine

Permutations by Cutting and Shuffling .....cccccccccccc0c0000000+. aL

Chapter 3 - Analyzing Shuffles..........cccssssssecssscesssoveseareres 45

DPIVINg NOT pS SHUI ANOIVSISU see eps ee 45

Chapter 4 - Exploiting Shufflles ............ssccssssscsssssereesscesees 66

Applying Hannum ’s Shuffling ANALYSIS .......2c00sc0evereere 66

ACE LOCALION Wsicucsc a eee eric nee, eae eee 79

4 Two-Pass ComboiShufflese. ee eres eee 85

Chapter 6 - Predicting ACES ..........:cssscssssseerecerssserssenesseees 89

Card PeGQiCtiOn caro ea ee coe eee 8&9

Chapter 8 - How Much to Bet.............cccssssssssssssssscsssssece 120

CCIE RCH Hates ter dene Ns cae 2 Wig.Cees. SLR aR 120

AD DCTUICES ¢cecessssorsecoceduessstncccesesoseonassoesvassausestcaase<sentesesss 130

PADDCHOIN A. SIM JUNG 1)

NUON soca scree aeta natn oe ee ac ccnassecceresecaedeaccs-ciccsencesieccsaaets 139

RGLEF ENCES 2. .50.20csceenscsveccnsccescesscstssaccsssanassdsansesneseseseroutess 146

(DABLE2-1., A-PERFECT SHUEFLLE c.c.cc-cccsccesesseeccnecesessececececceovestess 33

10 Tables and Figures

Tables and Figures 11

Michael Barnett, East Perth, Western Australia

Dr. David Bayer, Columbia University, New York

Dr. Persi Diaconis, Stanford University, California

Dr. William R. Eadington, University of Nevada, Reno

Dr. Solomon W. Golomb, Univ. of Southern California, Los Angeles

Dr. Robert C. Hannum, University of Denver, Colorado

John. K. King, Detroit, Michigan

“Marko,” The Learned Pig Magic eZine, Panama

John May, Dublin, Ireland

Rob McGarvey, Toronto, Canada

David Morse, Cranston, Rhode Island

David Nichols, Birmingham, Alabama

Darren Phillips, New Mexico State University, Las Cruces

Michael Porstmann, Dachau, Germany

David G. Schwartz, University of Nevada, Las Vegas

Frank Scoblete, Chicago, Illinois

Ralph D. Slocum, Warren of California, San Diego

Norman Wattenberger, www.qfit.com, New York

Bill Zender, www./astresortconsulting.com, Las Vegas, Nevada

Abigail Kamelhair taught art for 10 years before

Queen of Hearts Jack of Diamonds

© Directional Publishing, Inc., 2004 www.directionalart.co

14 About the Cover Art