Sequence Induction: Summary and Findings
The concept
No real-world shuffle fully randomizes a deck, so the pre-shuffle card order affects the post-shuffle expected value (EV) and optimal strategy. Hand shuffles (riffles, strips, cuts) and some mechanical shuffles preserve part of the original order.
Sequence induction goes one step further: the player shapes the pre-shuffle order rather than just observing it. Playing decisions such as hitting, standing, splitting or playing extra hands decide which cards reach the discard tray, and in what order. The theory identifies two potential benefits:
- Improved post-shuffle expectation. Favourable groupings, such as an ace next to a ten-value card, are set up so that some survive the shuffle.
- Predictive knowledge. A known pre-shuffle order plus a model of the shuffle lets the player estimate where specific cards will land.
The approach builds on established advantage-play (AP) techniques: shuffle tracking (following groups of cards through a shuffle), ace sequencing (predicting an ace's arrival from the card before it) and key-card location. The new element is deliberately creating favourable order through play. Earlier empirical work on casino shuffle non-randomness, such as Gwynn and Snyder's, measured ordinary shuffles and did not test player-induced manipulation. It therefore neither confirms nor rules out this approach.
Findings from simulation work
Simulation shows that surviving ace-ten adjacency produces a measurable player edge, from about −0.33% to −0.88% house edge depending on setup.
Testing used a Python blackjack simulator. The main ruleset was 8 decks; dealer stands on soft 17 (S17); double after split (DAS); double on any two cards; resplit to 4 hands; no resplitting aces; original bet only lost to a dealer blackjack; no surrender; blackjack pays 3:2. The baseline house edge was about 0.418% over 200 million hands, matching Wizard of Odds figures.
| Scenario | Setup | Result |
|---|---|---|
| Baseline, 8 decks | Random shuffle, 75% cut-card shoe | House edge ~0.418% |
| Forced pairing, 8 decks | 20% of aces kept next to a ten | House edge ~−0.329% (player advantage) |
| Forced pairing, 8 decks | 100% of aces kept next to a ten | House edge rose (unexpected; not fully explained) |
| One guaranteed pair, 1 deck | 1 of 4 aces always stays with a ten | House edge −0.876% (standard error 0.032%, 95% confidence interval −0.939% to −0.813%, 12 million hands) |
| Baseline, 1 deck | Random shuffle | House edge ~−0.03% |
| Riffle survival | Gilbert-Shannon-Reeds (GSR) model, 3 riffles | Planted ace-ten pair survives 12.55% vs 1.96% random |
| Boundary card | Bottom-of-pack card, 3 riffles | 12.5% chance of being next card vs 1.96%; ~+4.27% edge over a 6-hand window |
- Shoe model. An early model reset the shuffle after every hand and gave a misleading dealer-favouring result caused by the reset itself. Later testing used a 75% cut-card shoe (cards dealt to 75% penetration, then reshuffled).
- Passive edge. The single-deck −0.876% edge needs no observation, memorization or bet variation: a flat-betting basic-strategy player receives it. The practical question is how the correlation gets into the shoe, not how to extract it.
- Deck-count scaling. Single-deck effects were about 8 times larger than 8-deck equivalents, because each card is a far larger share of 52 cards than of 416.
- Riffle decay. Pair survival fell to the random level at about 7 riffles.
- Pack size. Shuffles needed for randomness grow roughly with the logarithm of pack size, so combining decks into a shoe adds less protection than intuition suggests.
- Earlier theoretical estimate. A model using Epstein's clustering probabilities for expert shufflers suggested that surviving ace-ten pairs could roughly double the blackjack rate in tracked zones, worth about 2% of additional edge. This was a calculation, not a simulation.
Open issues and limitations
The largest gap is calibration: the forcing levels tested are not tied to any measured level of player influence at a real table.
- Calibration. Forcing levels of 10%, 20%, 25% and 100% were tested. No published data shows how far a player can actually shift the probability of a ten following an ace.
- Unresolved modelling. Alternating deal order was not tested, and two plausible models gave opposite signs for the effect.
- Creating the pairing. The proposed method is to play several hands so a chosen pair lands in a known tray position. Untested: whether the casino's collection procedure keeps that grouping, whether the shuffle keeps it through to the deal, and the cost of extra minimum-stake hands while waiting.
- Recognition in multi-deck games. Spotting a specific tracked card after the shuffle works in single-deck games. It largely fails in an 8-deck shoe, which holds 128 ten-value cards and 8 copies of each card.
- General constraints. Thorough shuffles and continuous shuffling machines (CSMs) remove the effect. Deviating from basic strategy costs EV that must be recovered after the shuffle. Casinos can change shuffle procedures or bar suspected APs.
- Novelty. The method may not be covered in published AP literature, but that is separate from whether it works in practice.
Conclusion
The theory is sound: any finite shuffle leaves some pre-shuffle structure intact, and simulated ace-ten adjacency produces a measurable player edge. What remains unproven is whether a player can create that structure at a real table, at acceptable cost, against the shuffle procedure actually in use.