Professional Documents
Culture Documents
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
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.
PHOTOGRAPHIC CREDITS:
DISCLAIMER:
A CKNOWICASINENES: jc ccc.cc:ssaesnncecns
sebccecsasdacarsenacecereecsevanrdiunes 12
Table of Contents
Chapter 2 - Non-Random Shuffling ........ccccccscscsscsceeseeeees 32
Table of Contents
Chapter 5 - Locating ACES ........sssseseesesssssrsrenseerencensesensnees 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
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
Table of Contents 9
TABLES AND FIGURES
Thanks to:
“Ace”, www.ace-ten.com
12 Acknowledgments
Martin Joyal, Pierrefonds, Quebec, Canada
Abigail T. Kamelhair, Atlanta, Georgia
Acknowledgments
ABOUT THE COVER ART
Michael Dalton
Blackjack Review Network
www.BJRnet.com
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.
* 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.
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.
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
22 Ace Prediction
PUBLISHED REFERENCES
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.”
24 Ace Prediction
Unlocking the Shuffle: Trade Secrets of the Shuffle Trackers
Ace Prediction 25
How to Make $100,000 a Year Gambling
for a Living (1997)
26 Ace Prediction
Card Sequencing (2001)
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.”**
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.”
He had the upper hand, now all bets are off (2003)
Ace Prediction 29
R&R), and if you get the chance, pounce!””°
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
Ace Prediction 31
NON-RANDOM SHUFFLING
*
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
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
Non-Random Shuffling 33
Imperfect shuffle, also called the “amateur” shuffle
34 Non-Random Shuffling
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
Non-Random Shuffling 35
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:
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
* [FS] p. 188.
T FS] p. 186.
36 Non-Random Shuffling
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.”
Ey eed an
Where r = 1,2,3...
P(n) = (8/9)(1/9)""
Where 7=1);2,3:..
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:
38 Non-Random Shuffling
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).°°
1.00 ste
Non-Random Shuffling 39
Hannum’s results for one, two and three riffles are compared
in Table 2-6:
40 Non-Random Shuffling
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=]
I= log,(n!) - U 2-6
Where,
n=52
U=0
I= log,(52!) - 0 = 225.58 bits
Non-Random Shuffling 41
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).
42 Non-Random Shuffling
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.
Non-Random Shuffling 43
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.”’”*
44 Non-Random Shuffling
ANALYZING SHUFFLES
* [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.
46 Analyzing Shuffles
RESULTS
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.
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
* .
Based on a “perfect” riffle.
Analyzing Shuffles 47
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.
Where,
X0 (the assumed mean) = | card
A=-44
n= 1500
Where,
B=558
A’ = 1936
n= 1500
* [FS] p. 178.
48 Analyzing Shuffles
the confidence interval will contain the true population mean
LL.
Analyzing Shuffles 49
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
Where,
Xo (the assumed mean) = 3 cards
A151
n= 1440
Where,
B= 3249
A =3276)
n= 1440
50 Analyzing Shuffles
Sample standard deviation s = 1.50 cards
Sample size n = 1440
Confidence interval c = as / Vn where a = 1.96
Analyzing Shuffles 51
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
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.
X=xXot+A/n
= 7 + (-0.27) = 6.73 cards
Where,
*0 (the assumed mean) = 7 cards
A = -1208
n= 4400
52 Analyzing Shuffles
s=V((B - (47/n))/(n-1))
= V((41624 - (1459264 / 4400)) / (4399)) = 3.06 cards
Where,
B=41624
A’ = 1459264
n= 4400
Analyzing Shuffles 53
BROKEN SEQUENCES
54 Analyzing Shuffles
Where plugging the cutoffs, strip shuffling or boxing the deck
could cause sequence breaks, they also must be counted.
Analyzing Shuffles 55
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
56 Analyzing Shuffles
sequence is 0. The twice-riffled deck F+I is placed on the table
to form the final stack.
Analyzing Shuffles 57
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.
58 Analyzing Shuffles
Dealer grabs A and J. There are no ja
sequences in J. Chance of a broken T''e4
sequence is 0.
Sequence OR
GING? eae 4/511
c3,c4— 4/51
c5,c6 = 4/51
c7,c& 4/51
Total 16/51
60 Analyzing Shuffles
TABLE 3-6. BROKEN SEQUENCES BY COMPUTER
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
Analyzing Shuffles 61
“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:
62 Analyzing Shuffles
TABLE 3-7. EFFECT OF PLUGS, STRIPS AND BOX CUTS
Analyzing Shuffles 63
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:
AN UNCUT DIAMOND
64 Analyzing Shuffles
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.”
Analyzing Shuffles 65
EXPLOITING SHUFFLES
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
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.
Exploiting Shuffles 67
RESULTS
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
* [FS] p. 192.
68 Exploiting Shuffles
SIMPLE TESTS FOR RANDOMNESS
° Test*
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.
Exploiting Shuffles 69
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%:
Pre-Shuffle Position
Position 105-156 57-208 | 209-260 261-312
1% 21%
27%
8%
70 Exploiting Shuffles
Concordance among Shuffle Iterations
Where,
YR’ = squared sum of ranks of average post-shuffle positions
N=48
n=4
* [FS] p. 174.
+ [FS] pp. 200-201.
Exploiting Shuffles sp
MAPPING SHUFFLE WEAKNESSES
- Cn iceman e
oor * |
260 ey ¢
aad wee me
ee)
208 lo F Salat
7“ ¢ Kae
wer »
156 oes S cid oo
72 Exploiting Shuffles
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:
Exploiting Shuffles 73
nothing about the order of these cards, you will not be able to
“locate” them.
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.
74 Exploiting Shuffles
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:
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
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
Exploiting Shuffles 75
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
12194223 Xx 0 = 201
* [FS] p. 189
76 Exploiting Shuffles
probable values (95% prediction intervals) for post-shuffle
positions (Table 4-6):
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!).
78 Exploiting Shuffles
LOCATING ACES
ACE LOCATION
ACE SLUGS
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
0.000 Aces
Locating Aces 81
Probability of Consecutive Aces
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:
Values of P4., for two through seven Aces are in Table 5-3:
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)
* [FS] p. 186.
82 Locating Aces
A STEPLADDER SHUFFLE
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.
J+(A+G+D+K+B)
A+(G+D+K+B+H)
Final
84 Locating Aces
A TWO-PASS COMBO SHUFFLE
Locating Aces 85
Dealer grabs B and H. Cards are riffled together once.
Deck B+H (with two A4s) added to final stack.
Final
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.
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
* This 0.40 value is for perfect prediction of every card just prior to it being dealt.
Predicting Aces 89
THE PREDICTION PROBLEM
cl c2 3) c4
4 5 + 5 4 Ge +
4 4
q y e: o¢
A A B B
ell c3 mt 1
q e
f+| ‘s
A B
90 Predicting Aces
PA, 8B A ?
(A+B) 6-1
A 1
Dee tANB) =) (+l) = |
Nd
Predicting Aces 91
Key |
Ace a
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:
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:
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
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
] - r, or 1/9
94 Predicting Aces
r, OF 8/9
1 -r,or 1/9
mM
ie
BOX 3 BOX2 Box 1
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
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.
[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:
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
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):
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
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
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
im
Wide Space>| ¢ 4
+ q © Narrow Space
\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,
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
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.
HT
fober[4
BOXx4 BOXx3 BOX 2 BOXx1
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.
i"
Oe oe | Flee
in
ee
BOXx4 BOXx3 BOX 2 BOX 1
MATHEMATICAL EXPECTATION
Expected value for all first cards for one and four decks are
shown in Table 7-2:
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
* TFS] p. 161.
* [FS] p. 149.
MATHEMATICAL EXPECTATION
P(h)=a-(b+f) 7-2
Where,
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):
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
f=EN 7-4
Where,
J= fraction of bankroll
E(X) = expectation on the hand
v = variance on the hand
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.
119
Mathematical Expectation
How MUCH TO BET
EXPECTED RETURN
RISK
o=1.1Vn 8-2
Where, m = number of bets
Where,
k = fraction of Kelly Bet
j= expected return per unit bet = +0.04
y = variance per unit bet = o°= 1.21
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
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
k = fraction of Kelly Bet
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
HO Ge 8-5
Where,
a
k = fraction of Kelly Bet
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
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
x= s?/ ku 5-6
Where,
s = standard deviation per hour in bets
k = fraction of Kelly Bet
j4= expected return per hour in bets
Bankroll = US$40,000
Where,
r = exponential bankroll growth rate*
q = probability of doubling before halving
b=2
a=05
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
Appendices we
Three 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;
Appendices 133
MEAN OF D, (THREE-RIFFLE SHUFFLE)
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
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
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
146 References
1984, Chapter 3, “The Principles of the Game and Why It Can
Be Beaten,” p. 30.
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.
148 References
oe PETERSON, IVARS, “Card Shuffling Shenanigans,” Science
News, Vol. 162, No. 20, Saturday, 16 November, 2002.
** ETLING, LEAH, “He had the upper hand, now all bets are
off,” Santa Barbara News-Press, Sunday, November 9, 2003.
References 149
44 GOLOMB, SOLOMON W., “Permutations by Cutting and
Shuffling”, SIAM Review, Vol. 3, No. 4, October, 1961, p.
293:
150 References
- KNUTH, DONALD E., The Art of Computer Programming.
Reading, Massachusetts: Addison-Wesley Publishing
Company, Inc., 1969, pp. 124-151.
a , 1973, p. 845.
”” Ibid.
°8 GWYNN AND SNYDER, 1988, p. 7.
°" [bid.
6? GILBERT, EDGAR N., “Theory of shuffling,” Technical
Memorandum, Bell Laboratories, Murray Hill, New Jersey,
1955.
References 151
66 MAY, 2000,
p. 51.
™ FULVES,
1992, p. 114.
™ 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.
= , 1977, p. 179.
88 KONIK, 1995.
8°Ibid.
°° THORP, 1962, p. 98.
References 153
°4 BRAUN, JULIAN H., How to Play Winning Blackjack.
Chicago, Illinois: Data House Publishing Company, Inc.,
1980, pp. 82, 83.
ne , 1981, p. 93.
104
, Professional Blackjack. Las Vegas, Nevada: Pi
Yee Press 19945 pa245:
_ , 1994, p. 237.
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.
He 92004165,
ao Ibid.
116 , 2004, p. 5.
References 155
Blackjack Review Network
Gaining an edge in cards and life!
www.BJRnet.com
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
“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
FOREWORD BY
9"781879"712102
MICHAEL DALTON ISBN: 1-879712-10-5 $34.95 U.S «8Z9-
TH
TT