The One2Six Advantage-Play Project, Part 3: Designing the Blackjack Simulator and Experimental Baselines

Why the table and card flow had to be modelled before the shuffler

Part 1 defined the advantage-play question, while Part 2 used the available manuals and patents to reconstruct the One2Six as a plausible compartment-wheel mechanism. The next problem was deciding what needed to be built before that mechanism could be tested properly.

The easy approach would have been to simulate a continuous shuffler, notice a few unusual-looking sequences and begin constructing explanations around them. Blackjack supplies an endless amount of material for this kind of reasoning. Players remember clusters of dealer twenty-ones, repeated low cards and improbable runs far more readily than the thousands of ordinary hands surrounding them. Given enough output, a simulation can automate the same mistake at industrial scale.

The project therefore needed an experimental environment in which the One2Six could be compared with card-generating processes whose behaviour was already understood. That required a blackjack table capable of reproducing the physical movement of cards through a real game, because the machine receives its input from the table rather than from an abstract stream of independent draws.

From Mechanism to Experiment

Part 2 produced the following broad reconstruction:

ordered discards
    -> individual card feeding
    -> internal wheel compartments
    -> compartment storage
    -> output into a front shoe
    -> cards dealt back to the table

That description identified a plausible mechanism family, although it could not tell me whether the resulting output would contain anything useful. The machine would need to operate as one card source inside a controlled blackjack simulation, with the same rules, strategy and settlement logic applied to every source being compared.

The experimental sequence was therefore:

define the blackjack game
    -> define comparison card sources
    -> define the One2Six-style source
    -> run the same game against each source
    -> measure differences in the card stream
    -> determine whether any difference is visible to a player
    -> test whether the visible information has economic value

Each stage depends on the accuracy of the earlier ones. If the engine deals cards in the wrong order, returns discards too early or settles doubles incorrectly, those errors will flow directly into the One2Six results. The shuffler may be the object that motivated the project, but the table controls the batches of cards it receives and therefore forms part of the process being studied.

Why the Blackjack Engine Needs Physical Card Flow

Many blackjack simulators can safely ignore most of the physical life of a card. They draw a rank and suit, calculate the hand total, apply a strategy, settle the wager and continue. The discarded card may enter an abstract tray or disappear from the model until the next shuffle.

That level of detail works well for questions about basic strategy, rule variations or the long-run house edge of a conventional shoe. This project is concerned with the movement and eventual return of specific cards through a continuous mechanical system, which requires a much richer representation.

For every card dealt, the engine needs to retain enough information to determine:

• which physical card appeared;
• where it was dealt;
• when it left the table;
• where it entered the discard order;
• when the completed batch was returned;
• how the card moved through the selected source;
• when the same physical card appeared again.

The blackjack engine therefore creates the ordered card flow entering the machine. Any error in that flow changes the experiment before the shuffler has an opportunity to do anything.

The Central Architecture

The most important design decision was to separate card generation from the rules of blackjack. The game engine should request the next card without needing to know whether it came from an IID generator, a finite shoe, a manual casino shoe or a One2Six-style mechanism. Each source should expose the same basic interface for dealing cards and accepting returned discard batches.

The project was organised around the following layers:

Layer	Responsibility
Card model	Represents rank, suit, physical identity and individual draw events
Card source	Supplies cards and accepts returned discard batches
Game engine	Controls deal order, hands, boxes, legal actions and dealer play
Strategy policy	Chooses an action from those permitted by the game
Settlement layer	Calculates wagers, payouts and hand outcomes
Result tracker	Aggregates profit, event frequencies and streaks
Experiment layer	Runs controlled comparisons and produces diagnostics

This structure makes the card source the experimental variable. When two simulations use identical game rules, strategy and settlement logic while changing only the source, differences in the resulting card streams and game outcomes can be attributed to that source with much greater confidence.

It also prevents the One2Six assumptions from spreading through the blackjack engine. The machine can evolve as the reconstruction improves without requiring the rules, strategy and settlement code to be rewritten around it.

Baselines Before the Machine

The word random is often applied to several processes with very different properties. A useful experiment needs to distinguish those processes rather than combine them into one vague benchmark.

IID Random Cards

An independent and identically distributed source provides the cleanest mathematical baseline. Every draw is independent of the earlier draws, with the same probability assigned to each card symbol every time.

This does not reproduce the behaviour of a casino shoe because cards never become unavailable through depletion. Its value lies in the simplicity of its expected behaviour. Rank and suit frequencies, card recurrence and blackjack outcomes can be checked against known distributions. If the game engine produces unexplained structure under IID conditions, the fault is likely to be closer to the simulation than to the casino machine.

Finite Shuffled Shoe

A finite shoe introduces depletion by drawing from a fixed population without replacement. Every card removed changes the composition remaining, while discards stay unavailable until a reshuffle returns them.

This baseline allows the engine to work with persistent physical cards rather than newly generated symbols. It also provides the first comparison for recurrence and exclusion effects that arise naturally from a finite pack.

Manual Casino Shoe

The manual-shoe benchmark adds the procedural features of an ordinary casino game:

• multiple physical decks;
• dealing without replacement;
• a cut-card or penetration threshold;
• an ordered discard tray;
• no reshuffle during a round;
• reshuffling at the appropriate round boundary.

This provides the strongest conventional comparison because it represents the physical process that a continuous shuffler replaces. Any structure produced by the One2Six needs to be interpreted alongside the structure already created by a normal casino shoe.

Physical IID Recurrence

The mechanical-memory question also requires a memoryless baseline for the return of labelled physical cards.

A symbol-level IID generator can show whether another ten of spades appears, although a six-deck game contains six separate physical copies of that card. It cannot distinguish one copy returning from another copy of the same rank and suit being drawn.

A physical-IID source can solve that problem by assigning stable identities to 312 physical cards and selecting independently from the entire population on every draw. Each draw remains memoryless, while the identity of the selected physical card is retained for recurrence analysis.

The resulting baselines answer different questions:

Source	Main comparison
Symbol IID	Does the card-symbol stream behave independently?
Physical IID	Does a labelled physical card return as though every draw were memoryless?
Finite shoe	What structure arises naturally from dealing without replacement?
Manual shoe	What structure arises from ordinary casino shoe procedure?
One2Six-style source	What additional structure is created by the continuous mechanism?

A finite shoe will differ substantially from IID, and that difference is entirely expected. The useful comparison concerns the type of structure created by each process and whether the continuous mechanism introduces something beyond the familiar effects of depletion and casino procedure.

Physical Cards Need Two Identities

The card model requires a distinction between the physical object and each occasion on which that object is dealt.

physical_id
    stable identity of one physical card in the pack

draw_id
    unique identity of one deal event

Suppose a particular king of hearts is dealt in round 20, returned through the discard system and appears again in round 31. The two appearances share the same physical_id, while each receives a different draw_id.

This distinction supports several measurements central to the project:

• time between appearances of the same physical card;
• time from discard to first return;
• whether cards collected together return near one another;
• whether local discard order survives;
• whether individual cards return unusually quickly or slowly;
• whether compartment output creates groups of related cards.

Without stable physical identity, the simulator could still analyse rank and suit sequences, although the question of mechanical memory would largely disappear from the data.

The Discard Rack Forms Part of the Model

Cards do not return to the machine as an unordered collection. The game creates the discard rack through a sequence of actions and settlements. Busted hands may be removed immediately, while other player hands remain on the table until the round is settled. Splits create additional groups, player boxes are collected in table order and the dealer’s cards enter the rack separately.

The relevant flow is:

cards are dealt and played
    -> some hands are collected immediately
    -> remaining hands are settled
    -> player boxes are collected in order
    -> dealer cards are collected
    -> the completed rack is staged for return

The machine may erase that ordering so thoroughly that it has no measurable effect, but the ordering has to exist in the simulation before its survival can be tested.

The timing of the return also changes the available card population. Under the procedure being modelled, the completed rack from one round remains outside the card source during the next initial deal and is returned after that deal has been completed. Cards visible in the preceding rack are therefore excluded from the following initial deal because they have not yet re-entered the machine.

This creates a small player-observable exclusion window. Its economic value may eventually prove negligible, although returning the cards earlier in the simulation would remove a real feature of the operating procedure before it could be measured.

The Game Rules Still Matter

The One2Six source should handle cards, while the blackjack engine remains responsible for the rules of the game. The configured casino profile includes:

• initial dealing order;
• dealer drawing rules;
• blackjack settlement;
• doubles and additional wagers;
• splits and split-hand restrictions;
• timing of bust collection;
• dealer blackjack treatment;
• round completion.

These rules alter the number and order of cards entering the discard rack. A split consumes additional cards and creates separate hands. A double adds exactly one card to the player’s hand. A dealer drawing to a long total produces a different discard batch from a dealer standing immediately.

The card source ultimately receives the physical consequences of those rules. A realistic continuous-shuffler experiment therefore requires enough of the surrounding game to produce realistic input.

Strategy Must Remain Fixed During the Initial Comparison

The first comparisons between card sources should use one fixed playing strategy. That strategy provides consistent decisions under the selected casino rules and should not respond to information specific to the One2Six.

The clean comparison is:

same rules
same strategy
same wager
same settlement
different card source

Allowing the strategy to change between sources would make it difficult to separate the effect of the card process from the effect of the decisions being taken. A machine-aware strategy only becomes relevant after the project has established that the machine creates information a player could observe before acting.

The initial strategy therefore serves as a stable policy for generating comparable game outcomes. Its purpose at this stage is consistency rather than exploitation.

Source-Level Evidence Comes Before Profit

Profit is an attractive output because it appears to answer the practical question immediately. It also combines several separate sources of variation:

• card-source behaviour;
• blackjack rules;
• strategy quality;
• wager accounting;
• doubles and splits;
• serial dependence;
• ordinary blackjack variance.

A positive bankroll path can occur without an edge, while a real mechanical effect can remain hidden inside a noisy sequence of game results. Early experiments therefore need to examine the card source directly before moving to economic conclusions.

The most useful source-level measurements include:

Measurement	Question
Physical-card return time	How long before the same physical card appears again?
Discard-to-return latency	How long after collection before a card becomes visible again?
Same-batch reappearance	Do cards returned together emerge near one another?
Neighbour recurrence	Does local discard order survive the machine?
Rank and value autocorrelation	Does recent composition predict future composition?
Compartment ejection size	What groups does the mechanism deliver to the output?
Buffer occupancy	How much pre-existing output separates a returned card from the dealer?
Short-horizon composition	Do visible discards alter the expected upcoming cards?

Once those properties are understood, the analysis can move toward player blackjack frequency, dealer bust frequency, doubles, splits, win and loss distributions, conditional expected value and possible adaptive policies.

This ordering keeps the economic result connected to an identified mechanism. Otherwise, a favourable run could encourage a search for an explanation after the result had already been found.

What Would Count as Progress?

At this stage, progress did not depend on finding an edge. The immediate objective was to create an environment capable of testing the question honestly.

That required:

• a blackjack engine with correct card flow;
• interchangeable card sources;
• stable physical-card identities;
• an explicit discard order and return procedure;
• comparison sources with understood behaviour;
• configurable One2Six assumptions;
• experimental analysis kept separate from the game engine.

Once those pieces were in place, each card source could be tested without rebuilding the rest of the table. The significance of Part 3 was that the original question had developed into an experimental design detailed enough to implement.

What the One2Six Source Would Need

The next card source would represent the working mechanism reconstructed in Part 2:

ordered accepted discards
    -> feeder
    -> individual card insertion
    -> configurable carousel compartments
    -> compartment-level storage
    -> configurable output selection
    -> group ejection
    -> front output buffer
    -> dealer draw

Several parameters remained unresolved by the public evidence and therefore needed to remain visible within the design:

• compartment count;
• compartment capacity;
• card-insertion rule;
• order within a compartment;
• compartment-ejection rule;
• minimum ejection occupancy;
• output-buffer capacity;
• refill threshold;
• fallback behaviour;
• discard-return delay.

The design represents a family of plausible machines so that the assumptions can be varied. If an apparent effect disappears when one of those values changes, that sensitivity belongs in the result rather than being hidden inside a fixed configuration.

Where the Project Stood After Part 3

By the end of this stage, the project had a defined experimental structure. The mechanism from Part 2 had been translated into the requirements for a card source, while the surrounding blackjack engine had been designed around physical card movement, ordered discards and realistic return timing. The comparison sources provided separate baselines for symbol independence, physical recurrence, finite depletion and ordinary casino procedure.

The One2Six source itself had not yet been implemented, and the project had produced no advantage-play result. The design established the conditions under which any later result could be interpreted with reasonable confidence.

Part 4 moves into implementation by building the configurable One2Six-style source, moving physical cards through the carousel and output buffer, and verifying that the complete system can operate without losing, duplicating or inventing cards.

References

• mathematical-ev/shufflemaster-simulation. Public repository for the project, including the card model, card-source interfaces, blackjack engine and experimental framework developed through this stage.
• CARD one2six User Manual, 10 February 2005. Relevant to production procedure, including the front shoe, internal wheel, card inventory and the instruction to return discards after each hand.