Blackjack Math
APPLIED PROBABILITY INSTITUTE // QUANTITATIVE GLOSSARY

Blackjack Math Glossary

Authoritative mathematical definitions, closed-form formulas, and worked examples for 35 foundational concepts in blackjack probability and advantage play.

Basic Strategy & Decision EV

(11)

Basic Strategy

#
EV(action) = \sum_{s \in S} P(s) \cdot R(action, s)

The mathematically optimal decision policy (Hit, Stand, Double, Split, or Surrender) for every possible player hand against every dealer upcard, derived by exhaustive combinatorial backward induction assuming no card counting.

Worked Case Example: Standing on hard 12 against a dealer 4 has EV ≈ -0.223, whereas hitting has EV ≈ -0.229. Standing saves 0.6% in expectation over millions of rounds.

Expected Value (EV)

#
\mathbb{E}[X] = \sum_{i=1}^n x_i \cdot P(X = x_i)

The probability-weighted average payout of a decision or round over infinite trials. Expressed in units of initial bet.

Worked Case Example: Doubling 11 vs 6 in 6-deck S17 yields an EV of +0.672 units per bet, compared to +0.334 units for hitting.

Dealer Upcard

#
P(\text{Bust} \mid d \in \{5,6\}) \approx 0.42

The single card dealt face up to the dealer at the start of the round, serving as the sole Bayesian conditioning signal for player decision trees.

Worked Case Example: Dealer upcards of 5 and 6 present highest bust probabilities (42.6% and 42.3%), dictating conservative player standing.

Soft Hand

#
T_{\text{soft}} = A_{11} + \sum c_i \le 21

Any blackjack hand containing an Ace counted as 11 points without the total exceeding 21. A soft hand cannot bust on the immediate next hit.

Worked Case Example: Ace-6 is Soft 17. Basic strategy doubles soft 17 against dealer 3 through 6 to maximize EV against vulnerable dealer upcards.

Hard Hand

#
T_{\text{hard}} = \sum c_i \quad (A = 1 \text{ or no } A)

A hand that either contains no Aces or where all Aces must be counted as 1 point to avoid exceeding 21, carrying immediate risk of busting upon hitting.

Worked Case Example: A hand of 10-6 is hard 16, which carries a 61.5% probability of busting if hit with any card 6 through King.

Double Down

#
\text{EV}_{\text{double}} = 2 \cdot \left[ P(\text{Win}) - P(\text{Loss}) \right]

The option to double the initial wager in exchange for receiving exactly one additional card, providing high mathematical leverage on advantageous totals.

Worked Case Example: Doubling 10 against dealer 9 produces +0.38 EV per bet compared to +0.26 EV for hitting, capturing significant positive edge.

Pair Splitting

#
\text{EV}_{\text{split}} = 2 \cdot \mathbb{E}[V(c)] - \text{Covariance Term}

Dividing two cards of identical rank into two separate hands, each receiving its own independent wager and continuation play.

Worked Case Example: Splitting 8-8 against dealer 10 transforms an expected loss of -0.536 (playing 16) into an expected loss of -0.360 across two hands, saving 0.176 units.

Insurance Bet

#
\text{EV}_{\text{ins}} = 2 \cdot \frac{N_{10}}{N_{\text{rem}}} - 1 \cdot \frac{N_{\text{non-10}}}{N_{\text{rem}}}

A side bet of up to half the original wager offered when the dealer shows an Ace, paying 2:1 if the dealer holds a 10-value hole card.

Worked Case Example: Off the top of an 8-deck shoe, 10-density is 128/415 ≈ 30.84%. EV = 2(0.3084) - 0.6916 = -7.48%, making insurance a terrible bet without card counting.

Even Money

#
\text{EV}_{\text{take}} = +1.00, \quad \mathbb{E}[\text{refuse}] = 1.5 \cdot (1 - P_{10}) \approx +1.037

An option given to a player holding a natural blackjack when the dealer shows an Ace, guaranteeing a 1:1 payout instead of risking a push.

Worked Case Example: Refusing even money has an expected return of +1.037 units per hand. Taking even money forfeits 3.7% in long-term equity.

Dealer Bust Frequency

#
P(\text{Bust} \mid d) = \sum_{\text{paths}} P(\text{path}) \cdot \mathbb{I}(\text{Sum} > 21)

The probability that the dealer exceeds 21 conditioned on their initial upcard, spanning from 11.7% on Ace upcard to 42.6% on 5 upcard under S17.

Worked Case Example: Understanding dealer bust gradients explains why basic strategy stands on stiff totals (12–16) against upcards 2 through 6 but hits against 7 through Ace.

Hole Card

#
P(\text{Hole Card} = c) = \frac{1}{13} \quad \text{(Unconditioned)}

The dealer's face-down card in American-style blackjack games, which is checked for blackjack before player action when the upcard is a 10 or Ace.

Worked Case Example: Checking the hole card prevents players from losing split and double bets against an undiscovered dealer natural blackjack.

Card Counting & Indices

(13)

Running Count (RC)

#
\text{RC}_t = \sum_{k=1}^t w(c_k), \quad w \in \{-1, 0, +1\}

The raw cumulative algebraic sum of point values assigned to observed cards as they are dealt from the shoe.

Worked Case Example: In Hi-Lo, if three 5s (+1 each) and two Kings (-1 each) are dealt, the Running Count increments by +3 - 2 = +1.

True Count (TC)

#
\text{TC} = \frac{\text{RC}}{\text{Decks Remaining}}

The normalized running count per remaining unplayed deck. True Count normalizes the density of high cards across variable shoe penetration depths.

Worked Case Example: A running count of +6 with 2.0 decks remaining yields a True Count of +6 / 2.0 = +3.0, conferring an estimated +1.0% player edge.

Hi-Lo Counting System

#
w(c) = \begin{cases} +1 & c \in \{2,3,4,5,6\} \\ 0 & c \in \{7,8,9\} \\ -1 & c \in \{10,J,Q,K,A\} \end{cases}

The industry-standard Level 1 balanced card counting system developed by Harvey Dubner and Julian Braun, providing a betting correlation (BC) of 0.97.

Worked Case Example: Because small cards favor the dealer and tens/aces favor the player, positive Hi-Lo counts indicate that remaining shoe composition offers positive expected return.

Shoe Penetration

#
\text{Pen} = \frac{N_{\text{dealt}}}{N_{\text{total}}} \times 100\%

The percentage or fraction of the shoe dealt before the cut card triggers a reshuffle. Penetration governs count frequency and win rate variance.

Worked Case Example: In a 6-deck shoe (312 cards), dealing 4.5 decks (234 cards) equals 75% penetration. Shifting from 65% to 80% penetration more than doubles hourly EV.

Schlesinger's SCORE

#
\text{SCORE} = \frac{10^6}{N_0} = 10^6 \cdot \frac{\mu^2}{\sigma^2}

Standardized Comparison of Risk and Expectation (Don Schlesinger), defining hourly win rate in dollars per 100 hands on a normalized $10,000 bankroll at 13.53% risk of ruin.

Worked Case Example: A game with SCORE = 50 yields $50.00/100 hands under optimal Kelly wagering; games with SCORE below 20 are generally not worth playing.

Illustrious 18

#
\text{TC} \ge I_k \implies \text{Deviate from Basic Strategy}

The 18 highest-value basic strategy index deviations compiled by Don Schlesinger, capturing roughly 80%–85% of total playing efficiency gain.

Worked Case Example: Insurance at TC ≥ +3, 16 vs 10 standing at TC ≥ 0, and 12 vs 3 standing at TC ≥ +2 represent top Illustrious 18 plays.

Fab 4 Surrender Indices

#
\text{LS Index Cutoffs for 15 vs 9, 15 vs 10, 15 vs A, 14 vs 10}

The four most mathematically impactful Late Surrender index plays in card counting: 15 vs 10 (TC ≥ 0), 15 vs 9 (TC ≥ +2), 15 vs A (TC ≥ -1), and 14 vs 10 (TC ≥ +3).

Worked Case Example: Surrendering hard 14 against a 10 when the true count reaches +3 protects substantial equity compared to hitting into a ten-rich shoe.

Bet Spread

#
\text{Spread} = \frac{\text{Bet}_{\max}}{\text{Bet}_{\min}} \quad (1\!:\!8 \text{ to } 1\!:\!16)

The ratio between a card counter's maximum wager (at high positive True Counts) and minimum wager (at negative or neutral counts).

Worked Case Example: In a 6-deck shoe, a 1:12 spread ($25 min at TC ≤ +1 up to $300 max at TC ≥ +5) is required to generate a 1.2% net advantage.

Betting Correlation (BC)

#
\text{BC} = \text{Corr}(\Delta \text{EV}(c), w(c))

The statistical correlation between a counting system's point values and the actual change in player expectation produced by card removals. Hi-Lo achieves BC = 0.97.

Worked Case Example: A BC of 0.97 indicates that the count system tracks 97% of all potential theoretical betting gain available in the shoe.

Playing Efficiency (PE)

#
\text{PE} = \text{Corr}(\Delta \text{Strategy}, w(c))

The capability of a card counting system to predict optimal decision changes (standing, doubling, splitting). Level 1 Hi-Lo exhibits PE ≈ 0.51.

Worked Case Example: Multi-level counts like Wong Halves achieve PE = 0.57, but the cognitive burden often leads to operational execution errors.

Wonging (Back-Counting)

#
\text{Action: Enter at TC } \ge +1.5, \text{ Exit at TC } < 0

Technique popularized by Stanford Wong of counting cards from behind the table without betting, only entering the game when the count is advantageous (TC ≥ +1.5).

Worked Case Example: Wonging eliminates negative-expectation rounds entirely, raising player edge to +1.5% while drastically reducing bankroll variance.

Penetration Drag

#
\Delta \text{EV} \propto \text{Penetration}^2

The steep non-linear loss of expected hourly value that occurs when a casino cuts the shoe shallowly (e.g., placing cut card at 2.0 decks instead of 1.0 deck).

Worked Case Example: Shifting from 1.0-deck cutoff (83% penetration) to 2.0-deck cutoff (67% penetration) slashes a counter's hourly expectation by 58%.

Penetration Threshold

#
\text{Cut} \le 1.5 \text{ decks out of 6 (} \ge 75\% \text{)}

The minimum shoe depth required before a game becomes viable for positive-EV advantage play. 75% is the standard competitive cutoff in shoe games.

Worked Case Example: Games cutting off 2.0 or more decks in a 6-deck shoe fail the viability test and cannot be beaten profitably at acceptable variance.

House Edge & Casino Rules

(10)

House Edge (HE)

#
\text{HE} = -\frac{\mathbb{E}[\text{Net Profit}]}{\text{Initial Wager}} \times 100\%

The statistical ratio of expected casino profit to total initial wager under perfect basic strategy. The mathematical complement to Player RTP (RTP + HE = 100%).

Worked Case Example: Standard 6-deck S17 with DAS and Late Surrender exhibits a casino advantage of exactly 0.43% (Player RTP = 99.57%).

Double After Split (DAS)

#
\Delta \text{HE}_{\text{DAS}} \approx -0.14\%

Casino rule allowing players to double down on hands resulting from a pair split. Reduces casino house advantage by approximately 0.14%.

Worked Case Example: With DAS, splitting pairs like 2-2, 3-3, and 6-6 becomes mathematically favorable against weak dealer upcards due to potential doubling on resulting 9, 10, or 11.

Late Surrender (LS)

#
\text{Threshold: } \mathbb{E}[\text{Play}] < -0.50

The option to forfeit half the initial bet and end the hand after the dealer checks for blackjack. Reduces casino house advantage by ~0.08%.

Worked Case Example: Surrendering hard 16 vs dealer 9, 10, or Ace guarantees a loss of exactly -0.500, whereas hitting hard 16 vs 10 yields an inferior EV of -0.540.

S17 vs H17 Rule

#
\Delta \text{HE}_{\text{H17}} \approx +0.22\%

Table regulation governing dealer action on soft 17. S17 requires the dealer to stand on soft 17, while H17 forces a hit, shifting the casino advantage by +0.22%.

Worked Case Example: When the dealer hits soft 17, dealer bust rate rises slightly, but when completing hands they frequently draw to 18–21, tilting odds against the player.

Natural Blackjack (3:2)

#
P(\text{Natural}) = 2 \cdot \frac{4}{52} \cdot \frac{16}{51} \approx 4.83\% \quad (1\text{-Deck})

A two-card hand totaling 21 comprised of an Ace and a 10-value card, paying standard 3:2 (+150% return) unless the dealer also holds blackjack.

Worked Case Example: On a $100 bet, a 3:2 payout awards $150 profit. In an 8-deck shoe, natural blackjack frequency is approximately 4.75%.

6:5 Payout Trap

#
\Delta \text{HE} = +1.39\% \quad (6:5 \text{ vs } 3:2)

Deceptive casino rule mutation paying 6:5 ($120 on $100) instead of 3:2 ($150 on $100) for natural blackjacks, adding a devastating +1.39% to house edge.

Worked Case Example: A 6:5 payout completely obliterates basic strategy protection, shifting a 0.5% house edge game to an unplayable 1.9%–2.3% casino hold.

European No Hole Card (ENHC)

#
\Delta \text{HE}_{\text{ENHC}} \approx +0.11\%

Rule where the dealer receives no hole card until all player hands are completed. If the dealer makes a blackjack, players lose all doubled and split wagers.

Worked Case Example: Under ENHC, doubling 11 vs dealer 10 or Ace becomes significantly worse because a dealer blackjack claims the double wager.

Continuous Shuffling Machine (CSM)

#
\text{Penetration} \to 0, \quad \text{Composition} \approx \text{Constant}

An automated mechanical card dispenser that immediately inserts discarded cards back into the shoe, rendering traditional card counting ineffective.

Worked Case Example: CSMs slightly decrease the house edge by ~0.03% due to perpetual card replacement, but drastically increase hands per hour by 20%–30%, escalating casino hold.

Resplitting Aces (RSA)

#
\Delta \text{HE}_{\text{RSA}} \approx -0.08\%

Casino rule allowing players who receive another Ace after splitting Aces to split again up to 3 or 4 total hands. Decreases casino advantage by ~0.08%.

Worked Case Example: Because split Aces yield 21 frequently on a 10-card draw, RSA provides valuable extra high-EV action nodes.

Hit Split Aces

#
\Delta \text{HE} \approx -0.19\%

Rare player-favorable rule permitting multiple hits on hands created by splitting Aces, rather than receiving only one single card per Ace.

Worked Case Example: If dealt an Ace-2 on split Aces, a player can continue hitting to build a strong hand, cutting house edge by nearly 0.19%.

Bankroll, Variance & Risk of Ruin

(7)

Kelly Criterion

#
f^* = \frac{\mu}{\sigma^2} = \frac{\text{Player Edge}}{\text{Variance per Hand}}

The optimal capital allocation formula that maximizes the asymptotic growth rate of wealth while driving risk of ruin to zero over infinite trials.

Worked Case Example: With a 1.5% edge and variance of 1.32 per hand, full Kelly sizes the wager at 0.015 / 1.32 ≈ 1.14% of current bankroll.

Risk of Ruin (RoR)

#
\text{RoR} = \exp\left( -\frac{2 \mu B}{\sigma^2} \right)

The mathematical probability that a player's bankroll drops to zero before reaching a target growth threshold, derived from Brownian motion first-exit boundaries.

Worked Case Example: A bankroll of 500 units with win rate 0.01 units/hand and variance 1.3 has RoR = exp(-2 * 0.01 * 500 / 1.3) = exp(-7.69) ≈ 0.046%.

N-Zero Metric (N₀)

#
N_0 = \frac{\sigma^2}{\mu^2}

The number of hands required for expected cumulative earnings to equal exactly one standard deviation of variance, signaling emergence from statistical noise.

Worked Case Example: With expected profit μ = 0.015 units/hand and variance σ² = 1.33, N₀ = 1.33 / (0.015)² ≈ 5,911 hands (roughly 60 hours of play).

Split Hand Covariance

#
\sigma^2_{\text{split}} = \sigma_1^2 + \sigma_2^2 + 2 \cdot \text{Cov}(H_1, H_2)

The statistical measure of co-dependence between two hands produced by splitting a pair, sharing the same dealer outcome and initial card.

Worked Case Example: Because both split hands face the identical dealer outcome, their returns are positively correlated (Cov ≈ 0.28), significantly elevating single-round variance.

Standard Deviation per Hand (σ)

#
\sigma_{\text{hand}} \approx 1.15 \text{ to } 1.18 \text{ units}

The dispersion of individual hand outcomes around expected value. In blackjack, doubling and splitting inflate per-hand standard deviation to roughly 1.15 units.

Worked Case Example: Over a 1,000-hand session, total standard deviation is 1.15 * sqrt(1000) ≈ 36.4 units, meaning a 95% confidence interval spans ±71.3 units around EV.

Bankroll Requirement

#
B_{\text{req}} = \frac{-\sigma^2 \ln(\text{RoR})}{2 \mu}

The minimum capital reserve necessary to support a chosen betting spread and survive drawdown swings at a specified risk of ruin threshold (e.g. RoR ≤ 1%).

Worked Case Example: For a 1-to-12 spread with $25 minimum units, surviving downswings at 1% RoR typically requires a bankroll of 800 to 1,000 top units ($12,000–$15,000).

Drawdown Recovery Time

#
\mathbb{E}[T_{\text{recov}}] = \frac{D}{\mu}

The expected volume of hands required to recover from a peak-to-trough capital drawdown D under positive expectation μ.

Worked Case Example: A 100-unit downswing with an EV of 0.02 units/hand requires an average of 100 / 0.02 = 5,000 hands of grinding to reach a new all-time high.

Combinatorics & Hypergeometric Models

(4)

Hypergeometric Distribution

#
P(X = k) = \frac{\binom{K}{k} \binom{N-K}{n-k}}{\binom{N}{n}}

The discrete probability distribution describing sampling without replacement from a finite shoe of cards, governing all non-replacement card draws.

Worked Case Example: Calculating the exact probability of drawing two 10s from a depleted 6-deck shoe requires hypergeometric combinations rather than binomial approximation.

Central Limit Theorem Convergence

#
\lim_{n \to \infty} P\left( \frac{S_n - n\mu}{\sigma \sqrt{n}} \le z \right) = \Phi(z)

The statistical law establishing that cumulative blackjack profits asymptotically follow a Gaussian normal distribution as sample size n exceeds several thousand hands.

Worked Case Example: While individual hand returns are discrete (-2, -1, 0, +1, +1.5, +2), the sum of 10,000 rounds forms a smooth bell curve centered at n * EV.

Cut Card Effect

#
\mathbb{E}[\Delta \text{EV}_{\text{cut}}] \approx -0.01\% \text{ to } -0.02\%

The subtle statistical bias caused by dealing until a physical cut card appears, slightly favoring the casino because rich shoes consume cards faster and reach the cut card sooner.

Worked Case Example: First analyzed mathematically by Gordon and Thalheimer (1992), this effect imposes a tiny -0.015% drag on shoe expectations.

Burn Card

#
N_{\text{unseen}} = N_{\text{total}} - N_{\text{burned}} - N_{\text{played}}

Card removed unseen and placed into the discard tray prior to dealing the first round after a shuffle to deter card marking.

Worked Case Example: Because burn cards are selected at random without replacement from the freshly shuffled shoe, they do not bias expectations in either direction.