Optimizing Expected Value in Blackjack Basic Strategy

Abstract

A player who has memorized a basic strategy chart knows less about blackjack than the chart implies. The chart is the endpoint of a calculation, not the calculation itself — and treating it as a finished answer rather than a compressed decision policy leads to exactly the kind of mistakes it was designed to prevent: applying an S17 chart to an H17 game, standing on soft 18 because eighteen is “already a good hand,” or sitting down at a single-deck table without noticing the 6:5 printed on the felt.

Under fixed rules, basic strategy is the operational form of an expected-value maximization problem defined over a finite set of legal player decisions. Roger Baldwin and his colleagues established this in 1956, framing optimal blackjack play as a problem of maximizing mathematical expectation rather than following intuition. This paper treats blackjack as a finite, rule-constrained decision process governed by sampling without replacement, conditional probability, deterministic dealer behavior, and asymmetric payoff structures. For any specified rule set — number of decks, blackjack payout, soft-17 policy, doubling permissions, splitting restrictions, surrender availability, and peek procedure — the correct player action at each decision state is the legal action with the greatest expected value.

The exact solution is composition-dependent. It depends on the precise cards already visible, not merely on the player total and dealer upcard — a distinction Peter Griffin developed rigorously in The Theory of Blackjack. Total-dependent basic strategy is a workable compression of that richer policy, but the underlying criterion in both cases is the same: maximize expected value at every decision state.

Several counterintuitive plays — hitting hard 16 against a ten, splitting 8s against a ten, surrendering, doubling soft totals against weak dealer upcards — have precise expected-value justifications that conflict sharply with what casual players believe “safe” or “smart” play looks like. Basic strategy does not guarantee profit, reduce variance, or overcome a negative-expectation rule environment. It prevents avoidable loss by selecting the best available decision at each state.

Keywords: blackjack, basic strategy, expected value, combinatorial probability, finite decision process, casino mathematics, dependent probability, dynamic programming

I. Introduction

Spend an afternoon at any blackjack table and you will hear the same handful of explanations for why players made the decisions they did. Never take a card when you might bust. Assume the dealer has a ten underneath. Let the dealer break. Some of these coincide with correct strategy in limited situations. None of them constitutes a decision rule.

The actual standard is expected value — a concept most players have heard of but few apply with any rigor. Every legal player action produces a probability-weighted distribution of payoffs. The correct decision is whichever action produces the highest expected value among the available alternatives. That standard does not care how a decision feels, whether it worked out last time, or whether a player across the table just busted on the same card. It measures the decision against all possible outcomes weighted by their likelihood.

Roger Baldwin, Wilbert Cantey, Herbert Maisel, and James McDermott made this explicit in 1956, publishing the first rigorous mathematical treatment of blackjack strategy in the Journal of the American Statistical Association. They framed the problem as maximizing mathematical expectation, not following gambling tradition. Ed Thorp extended that work in Beat the Dealer, demonstrating both the theoretical and practical consequences of the framework. Peter Griffin, in The Theory of Blackjack, pushed the analysis further into composition-dependent strategy, expected values at specific hand states, and the theoretical basis for counting systems. Don Schlesinger’s Blackjack Attack translated the advanced mathematics into a form that players could actually apply. The mathematics has been settled for decades. The way it is usually communicated has not been.

For any fixed blackjack rule set and specified information structure, the optimal player decision at each legal state is the action that maximizes expected value across all possible future card sequences and payoffs. That is the formal content of basic strategy — the chart is where the math ends up after that calculation is complete, not a set of safe habits, not conventional wisdom, not a shortcut that bypasses the need for understanding.

II. Rule Conditions and the Need for Specification

No blackjack analysis means anything unless the rules have been specified first. This is not a methodological nicety — it is the whole foundation of the analysis. Blackjack is not one game. It is a family of related games whose expected values shift, sometimes dramatically, when the rules shift.

The rule vector for any given game can be written as:

R = (d, b, S17/H17, DAS, RSA, SPL, DD, LS, PEEK, CSM)

where d is the number of decks; b is the blackjack payout; S17/H17 indicates whether the dealer stands or hits soft 17; DAS indicates whether doubling after splitting is permitted; RSA indicates whether aces may be resplit; SPL describes split and resplit limits; DD describes doubling restrictions; LS indicates late surrender availability; PEEK describes the dealer blackjack-checking procedure; and CSM indicates continuous shuffling.

Every element of that vector affects either which actions are legal, how those actions are valued, or both. A six-deck H17 game requires a different strategy than a double-deck S17 game — not a tweaked version of the same strategy, but a meaningfully different one across several decision states. A game with late surrender available requires different decisions than one without it. As Robert Hannum and Anthony Cabot detail in Practical Casino Math, these rule interactions determine the house edge and the player’s correct response to each hand state.

Single-deck 6:5 games illustrate this clearly. Players associate fewer decks with better conditions — and in a 3:2 game, that association is correct. But the reduced blackjack payout in a 6:5 game removes so much value from the player’s best hands that a single-deck 6:5 game can be worse than a six-deck 3:2 game, something Don Schlesinger quantifies in Blackjack Attack. The “Single Deck” sign attracts players who have heard that single-deck games are favorable. The 6:5 on the felt, often printed small, is the actual game they are sitting down to play. From the floor, the pattern is consistent: players read the banner, not the payoff.

III. Information Structure: Total-Dependent and Composition-Dependent Play

A standard basic-strategy chart classifies the player’s hand by total, softness, pair status, and dealer upcard. That information set can be written formally as:

I_T = (player total, hard/soft/pair status, dealer upcard, R)

The mathematically exact policy uses more. It distinguishes between hands with the same total but different card composition, and accounts for the changing composition of the remaining shoe:

I_C = (H, u, C, R)

where H is the exact player hand, u is the dealer upcard, C is the remaining shoe composition, and R is the rule vector.

The difference exists because blackjack is dealt without replacement. A hard 16 composed of 10+6 is not compositionally identical to one composed of 9+7, 8+5+3, or 4+4+4+4. These hands share a total but remove different ranks from the shoe, and the expected values of hitting, standing, doubling, or surrendering can differ as a result — a point Peter Griffin traces through in considerable detail in The Theory of Blackjack.

For ordinary play at a fixed bet, total-dependent strategy is the practical standard. It is learnable and captures nearly all of the decision value available to a non-counting player. Composition-dependent deviations matter most in single-deck games, where the removal of a few cards shifts the remaining distribution meaningfully. In a six-deck shoe, the differences are smaller. The total-dependent chart is an approximation — a very good one for most practical purposes, but an approximation nonetheless.

IV. Blackjack as a Finite Probability Space

A blackjack round is a finite probability space. The shoe contains a fixed number of cards. The player has a finite set of legal actions at each decision point. Casino rules bound the depth of the decision tree by limiting splits, resplits, doubles, and draws to split aces. The dealer follows a deterministic rule. From any state, the game tree terminates in a win, loss, push, blackjack payoff, surrender, or bust.

The remaining shoe composition at any point is:

C = (c_A, c_2, c_3, ..., c_9, c_10)

where c_10 represents tens, jacks, queens, and kings collectively. In a fresh six-deck shoe:

C_0 = (24, 24, 24, 24, 24, 24, 24, 24, 24, 96)

The probability of drawing rank r from the current composition is:

P(r | C) = c_r / Σ_j c_j

After that card is drawn, the composition updates:

C' = C - e_r

where e_r is the unit vector for rank r.

Roulette wheels do not lose numbers after a spin. Dice retain all faces between rolls. In blackjack, every card exposed — whether in a player’s hand, the dealer’s hand, or a burn — changes the composition of what remains, and that composition determines the probabilities that govern every subsequent decision. Stewart Ethier develops this formally in The Doctrine of Chances. At the table, the practical consequence is that informal intuitions built on the assumption of independent trials misrepresent the structure of the game entirely. The shoe has no memory in the sense of carrying an obligation — but its composition carries information, and that information shifts with every card dealt.

V. Dealer Policy and Terminal Outcomes

The dealer is executing a rule. There is no judgment involved, no read of the table, no assessment of what the players are holding. In an H17 game, the dealer hits all totals of 16 or less and also hits soft 17. In an S17 game, the dealer stands on all totals of 17 or more, soft included. Robert Hannum and Anthony Cabot cover the implications in Practical Casino Math.

The probability that the dealer reaches terminal outcome t, given dealer upcard u, remaining shoe composition C, and rule vector R, is:

D_R(t | u, C)

where terminal outcomes are:

t ∈ {17, 18, 19, 20, 21, Bust}

A recursive model of dealer behavior gives:

D_R(d, C) = δ_d,                                        if d is terminal under R
           = Σ_r (c_r / N) D_R(f(d,r), C - e_r),       otherwise

where f(d, r) updates the dealer state after drawing rank r, N = Σ_j c_j, and δ_d is the terminal distribution.

Every player decision resolves against this distribution. When the player stands, the hand resolves against it directly. When the player hits or doubles and does not bust, the resulting hand still faces the same forced dealer process. The player has no influence over what the dealer does after acting — only over the state of their own hand when that resolution occurs.

VI. Expected Value and the Decision Criterion

Let X represent the player’s net result in units of the original wager. A win pays +1, a loss −1, a push 0. A natural blackjack pays +1.5 in a 3:2 game and +1.2 in a 6:5 game. Doubled hands scale both gains and losses by two.

For any legal action a at state S:

EV(S, a) = Σ_i P_i O_i

where P_i is the probability of outcome i and O_i is the associated payoff. The optimal decision is:

π*(S) = arg max_{a ∈ A(S)} EV(S, a)

and the value of the state is:

V(S) = max_{a ∈ A(S)} EV(S, a)

The correct play is the action with the highest expected value among legal alternatives — not the one most likely to win this particular hand, and not the one that avoids the most visible immediate risk. As Roger Baldwin’s 1956 paper established, and as Peter Griffin later formalized, the evaluation happens at the level of probability-weighted outcomes across the full distribution of possible results, not at the level of the next card.

VII. The Value of Player Actions

Standing

If the player stands with total p, the expected value is:

EV_stand(S) = m Σ_t D_R(t | u, C) g(p, t)

where m is the wager multiplier and:

g(p, t) = +1    if dealer busts or p > t
         =  0    if p = t
         = −1    if p < t

Standing is valuable when the dealer’s upcard creates enough bust probability or weak terminal outcomes to compensate for the player not drawing. Against a strong dealer upcard, that compensation often does not exist — the dealer’s terminal distribution is weighted toward high values, and the player’s standing total simply loses to most of them.

Hitting

Hitting is recursive. The player draws one card, and the hand either busts or enters another decision state:

EV_hit(S) = Σ_r (c_r / N) { −m,        if H + r busts
                              V(S_r),    otherwise }

where S_r is the successor state after drawing rank r. The value of hitting includes the expected value of all subsequent optimal decisions in the non-busting branches — not just the probability of busting on the next card. Players who evaluate hitting purely by bust risk are solving the wrong problem.

Doubling

Doubling increases the wager and limits the player to one additional card:

EV_double(S) = Σ_r (c_r / N) { −2,                        if H + r busts
                                 EV_stand(S_r; m=2),        otherwise }

Doubling is correct when the expected gain from increasing the wager outweighs the expected cost of surrendering future decision flexibility, as Don Schlesinger demonstrates in Blackjack Attack.

Surrender

Late surrender, where available, produces a fixed expectation of −0.5. It becomes correct when hitting, standing, and doubling all produce worse results — which happens in specific hand states, particularly weak totals against strong dealer upcards. Surrender does not improve the odds of winning the hand. At the table, it looks like giving up. In the expected-value calculation, it is sometimes the least irrational choice available.

Splitting

Splitting replaces a pair (p, p) with two hands beginning with p, subject to rule restrictions on resplitting, drawing, and doubling. Payoffs across the two resulting hands are additive in expectation:

E[X_1 + X_2] = E[X_1] + E[X_2]

Both hands share dependence on the same dealer outcome, which affects variance:

Var(X_1 + X_2) = Var(X_1) + Var(X_2) + 2 Cov(X_1, X_2)

John Nairn’s work on exact splitting calculations confirms that the additivity of expectation — not independence — is what makes the analysis tractable. Splitting can increase the total amount wagered on a given round, but that is a consequence of the math, not an argument against the play.

VIII. Basic Strategy as an Optimal Policy

The game tree is finite. Each legal action produces a finite distribution of terminal payoffs. Since the action set is finite, the set {EV(S, a) : a ∈ A(S)} has a maximum, and any action achieving that maximum is optimal:

π*(S) = arg max_{a ∈ A(S)} EV(S, a)

Terminal nodes have known payoffs. Non-terminal nodes inherit value through probability-weighted successors. Backward induction yields an optimal action at every reachable state. This is what Roger Baldwin’s group proved in 1956, and what Richard Epstein formalized in The Theory of Gambling and Statistical Logic: basic strategy is the practical surface of a finite expected-value optimization, not folk wisdom organized into a table.

IX. Why Correct Play Can Still Be a Losing Play

Watch a blackjack table long enough — not from a chair but from the pit — and one hesitation recurs more than any other. A player draws to a hard 16 against a dealer ten, stares at the cards for several seconds, and either stands because “I can’t take the risk” or hits with a grimace because “the book says so.” What the player rarely grasps in either case is that neither option is good. Both have negative expectation. The question is which one loses less.

Players expect “correct” to mean “likely to win.” In blackjack, it often means “least damaging,” and that gap between what players expect the chart to do and what it actually does is the source of most of the resistance to following it.

Hard 16 against a dealer ten is the standard case. Standing avoids an immediate bust. Hitting risks one. But standing leaves a weak total against one of the dealer’s strongest positions, and the hand resolves against a terminal distribution weighted heavily toward 17 through 21. Hitting introduces bust risk while also opening the possibility of improving to a total that can actually win.

If the expectation of hitting is −0.535 and the expectation of standing is −0.540:

−0.535 > −0.540

The margin is not dramatic. It does not need to be. The decision just needs to go to the option with the better number, even when both numbers are negative. As Don Schlesinger shows in Blackjack Attack, these marginal improvements compound across hundreds of decisions into a meaningful difference in long-run performance.

X. Hard Totals and the Stiff-Hand Problem

Hard 12 through 16 — stiff hands — are where players most frequently depart from correct strategy, and usually in one direction: they stand on everything because “I don’t want to bust,” regardless of what the dealer is showing. This plays out at tables across the industry, across cultures, across jurisdictions. The instinct is universal. So is its cost.

The correct treatment depends on two forces in opposition: the player’s bust probability on the next card, and the dealer’s probability of producing a weak or busted final total. Against dealer 4, 5, and 6, the dealer draws from many unstable intermediate totals under forced rules. Standing on a stiff hand can be correct not because the player’s hand is strong, but because the dealer’s structural position is weak enough that waiting costs less than acting. The player does not need a good hand — they need the dealer to fail, and the probabilities support that outcome often enough to justify the stance.

Against dealer 9, 10, or ace, the calculation reverses. The dealer’s terminal distribution shifts toward higher values, and the player standing on 12 through 16 wins primarily when the dealer busts. When that bust probability is too low to compensate, hitting becomes superior despite the immediate bust risk. Peter Griffin’s analysis in The Theory of Blackjack maps this boundary precisely for each upcard.

The “never bust” player loses more than necessary against strong dealer upcards because they are avoiding a risk that the math has already priced in.

XI. Soft Totals and the Value of Flexibility

Soft hands work differently because the ace can count as 11 or 1. Drawing a single card to a soft total almost never produces an immediate bust, and that flexibility changes the value of aggressive play in ways that players consistently underestimate.

Soft 18 is the hand where this goes wrong most often. A player sitting on soft 18 against a dealer 6 is frequently reluctant to double, reasoning that 18 is already a reasonable result. But the question is not whether 18 is decent in isolation — it is whether the expected value of doubling, given the dealer’s vulnerability at that upcard, exceeds the expected value of standing. Against a 6, the dealer draws from a weak position under forced rules, and the gain from increasing the wager on a flexible hand more than compensates for the loss of future drawing options, as Don Schlesinger details in Blackjack Attack.

Against a dealer ten, the same hand may call for hitting rather than standing, because 18 does not perform well enough against the dealer’s improved terminal distribution. The player is not trying to improve on a “good” hand — they are responding to a changed probability environment.

Soft-hand strategy is more aggressive than most players expect, and that aggression has a precise basis: the ace absorbs hitting risk in a way that hard totals simply cannot.

XII. Splitting Pairs and Rebuilding the Decision Tree

Because splitting sometimes requires putting more money into a hand that already looks weak, it is the decision type that generates the most visible resistance at the table. Players who will follow the chart on a hit or stand will override it on a split, and the override is almost always in the direction of not splitting when they should.

Splitting 8s against a dealer ten draws the most consistent objections. A pair of 8s is hard 16 — one of the worst non-bust totals in the game. Played as a single total, the hand is a losing proposition whether the player hits or stands. Splitting does not change that in any fundamental sense; it replaces one bad position with two less-bad ones. An 8 starting a new hand is more flexible than hard 16: a ten makes 18, an ace makes 19, a 3 makes 11, a 2 makes 10. None of those are guaranteed wins. The point is that the combined expected value of two hands starting on 8, played optimally, exceeds the expected value of playing the original 16, and that difference justifies the additional exposure. John Nairn’s exact calculations on pair splitting confirm the arithmetic. What they cannot resolve is the player’s reluctance to act on it when it means more money at risk against a dealer ten.

The opposite case — why splitting tens is wrong — is easier for players to understand intuitively but harder to get them to apply correctly in the other direction. Twenty is a strong made hand. Breaking it up replaces a high-probability winning position with two uncertain starting points, and the possibility of building two strong hands does not compensate for the expected-value loss of abandoning 20. Both examples show the same underlying principle: the value of a hand is not the current total on the felt. It is the expected value of the decision tree that follows from the actions available at that state.

XIII. Rule Changes and Their Mathematical Consequences

Blackjack Pays 3:2 or 6:5

The gap between 3:2 and 6:5 is 0.3 units per natural. From a floor perspective, players who notice the difference typically do so only after sitting down, and even then, many underestimate its significance on the grounds that it applies only to blackjacks. That is precisely why underestimating it is a mistake.

Naturals are among the best outcomes a player can receive. Reducing their payoff removes value directly from the hands that most favor the player, and no combination of other favorable rules reliably makes up for it. Don Schlesinger’s analysis in Blackjack Attack and Robert Hannum and Anthony Cabot’s work in Practical Casino Math both quantify the damage. A 6:5 blackjack game is a materially inferior game, and the single-deck version of it is not redeemed by the single deck. The label and the payoff are separate facts, and only one of them determines the edge.

Late Surrender

Late surrender converts any hand into a fixed expected value of −0.5. Against certain combinations — a stiff total against a dealer nine, ten, or ace — that fixed loss is better than the expected loss from playing the hand out. The point is not to avoid losing the hand; the hand is already likely to be lost. The point is to lose less than the alternatives would cost.

As Don Schlesinger documents in Blackjack Attack, the hands where surrender is correct are specific and limited. The value of its availability lies not in using it constantly but in having it as a rational option in the states where the alternative is a larger expected negative.

Dealer Hits Soft 17

H17 changes the dealer’s terminal distribution. Hitting soft 17 does allow additional bust opportunities the dealer would not have under S17, but it also allows the dealer to improve to 18, 19, 20, and 21 from positions that would have resolved as 17. The net effect is unfavorable to the player, as Robert Hannum and Anthony Cabot explain in Practical Casino Math. Because the dealer’s terminal distribution sits underneath every unresolved player decision, H17 affects not just one line of the strategy chart but several — standing decisions, soft doubling, and some split situations all shift when the dealer’s behavior changes.

Continuous Shuffling Machines

In a standard shoe game, cards dealt during a round are out of play until the shuffle point. In a CSM game, those cards cycle back into the randomization process almost immediately, reducing the influence of observed cards on the remaining composition and weakening whatever advantage a player might derive from tracking shoe depletion. James Grosjean addresses the composition implications in Beyond Counting.

The operational effect is the one most worth noting from a floor management perspective. Continuous shuffling removes idle time between rounds, which increases hands per hour. Since:

Expected Loss per Hour = Average Bet × House Edge × Hands per Hour

a higher pace increases hourly exposure even when the per-hand edge appears unchanged. The game moves faster, which looks like a benefit to players seeking action and functions as additional throughput for the house.

XIV. Variance and the Limits of Basic Strategy

Basic strategy does not smooth out blackjack. A player following it correctly will still have sessions that go badly wrong, and another player ignoring it will have sessions that go well. Neither outcome is a test of whether the strategy is correct — the sample size of a single session, or even a month of sessions, is far too small to reveal anything meaningful about decision quality.

Let X_1, X_2, …, X_n be the player’s results over n hands, with sample average:

X̄_n = (1/n) Σ_{i=1}^{n} X_i

Under stable rules and a fixed strategy, the law of large numbers implies:

X̄_n → E[X]  as  n → ∞

as Stewart Ethier establishes in The Doctrine of Chances. But convergence is slow relative to the scale of most gambling sessions. Blackjack generates substantial variance through naturals, doubles, splits, busts, and streaks that can run for extended periods in either direction. Short-run results are driven far more by that variance than by the quality of decisions behind them.

Basic strategy applies to one decision at a time. It says nothing about how large the wager should be, how long to play, how to manage a bankroll, or how to interpret a losing run. Those are separate problems, and confusing them with basic strategy leads to a predictable error: assuming that because a correct play lost, the play was wrong. The outcome was within the normal variance of a negative-expectation game played with minimum avoidable loss. That is different from the play being mistaken.

XV. Practical Implications for Players and Operators

Two things follow from this analysis for players. First, the strategy must match the actual rules in play. The correct chart for an S17 game is not the correct chart for an H17 game. A game with late surrender has decision states that simply do not exist in a game without it, and applying the wrong chart to either is not conservative — it is inaccurate, with an expected cost attached to that inaccuracy.

Second, emotional comfort is not a reliable decision criterion. Standing on a stiff total because hitting “feels too risky” is a real cost. Refusing to split when the math supports splitting is a real cost. Declining to surrender because it feels like admitting defeat is a real cost. These costs are individually small and collectively significant across thousands of decisions. Watching players absorb those costs over years, across tables in different countries, makes the pattern hard to miss: the emotional accounting and the mathematical accounting are measuring different things, and only one of them is keeping an accurate score.

For operators, the structure of the game is worth understanding on its own terms. The house edge is built into the rules and payoff structure — the player acts first and can bust before the dealer plays, the blackjack payout is asymmetric, and dealer behavior is fixed by rule. Player mistakes accelerate the math but are not required to maintain it. As Robert Hannum and Anthony Cabot document in Practical Casino Math, the actual edge is a product of the full rule vector, and the most accurate representation of any game’s economics comes from analyzing the complete set of rules — H17 versus S17, 3:2 versus 6:5, penetration depth, CSM versus shuffled shoe — not from highlighting whichever single rule appears most favorable on a sign.

XVI. Scope and Limitations

This paper addresses basic strategy under fixed rules and ordinary player information. Card counting, shuffle tracking, ace sequencing, hole-card play, side bets, promotional overlays, bankroll optimization, Kelly betting, risk of ruin, team play, procedural vulnerabilities, and advantage-play techniques are all separate subjects requiring different models.

Card counting expands the information state beyond what the total-dependent chart uses, incorporating estimates of the remaining shoe composition to adjust both decisions and bet sizing. Betting optimization changes the objective from maximizing hand expectation to maximizing bankroll growth or utility over time. Side bets have their own probability structures, independent of the base game calculation. Ed Thorp’s Beat the Dealer and Peter Griffin’s The Theory of Blackjack address these extensions in depth.

The result here is bounded to a specific claim: under a specified rule set and information structure, basic strategy is the expected-value-optimal decision policy at each hand state. It is not a claim that flat-betting blackjack under ordinary rules is a profitable activity.

XVII. Conclusion

After years at the table — watching players make decisions, watching dealers execute rules, watching the same mistakes surface in casinos across multiple continents — what stands out is not how often players choose the wrong action. It is how confident they are that their reasoning is sound. The player who stands on hard 16 against a ten because they “didn’t want to bust” has decided that avoiding an immediate visible risk outweighs the expected value of the alternative. The player who refuses to split 8s because “I’m not putting more money in with this dealer” has confused emotional accounting with mathematical analysis. These are understandable errors. They are also costly ones, and they persist because the chart is usually presented as a list of answers without any explanation of the problem it solves.

The mathematical structure behind basic strategy is not complicated to state, even if the computation behind it is demanding. The shoe is finite. The dealer follows a fixed rule. The player has a finite set of legal actions at each decision point. Each action produces a probability-weighted distribution of payoffs. The best action is the one with the highest expected value among the alternatives — a point Roger Baldwin and his colleagues established in 1956 and one that remains unchanged regardless of what happened on the last hand or the last session.

The chart is the compression of that calculation into a usable form. Understanding what the calculation is — and why it produces the answers it does — is what separates a player who follows the chart from one who actually knows the game.

Selected References

Baldwin, R. R., Cantey, W. E., Maisel, H., & McDermott, J. P. “The Optimum Strategy in Blackjack.” Journal of the American Statistical Association, 1956.

Epstein, R. A. The Theory of Gambling and Statistical Logic.

Ethier, S. N. The Doctrine of Chances: Probabilistic Aspects of Gambling.

Griffin, P. A. The Theory of Blackjack.

Grosjean, J. Beyond Counting.

Hannum, R. C., & Cabot, A. N. Practical Casino Math.

Nairn, J. A. “Exact Calculation of Expected Values for Splitting Pairs in Blackjack.”

Schlesinger, D. Blackjack Attack: Playing the Pros’ Way.

Thorp, E. O. Beat the Dealer.