Why shuffling matters to the maths
Every deck calculation assumes that each possible ordering of the deck is equally likely. A standard 52-card deck can be arranged in 52 factorial ways, a number with 68 digits, roughly 8 times 10 to the power of 67. If the shuffle is poor, some arrangements are more likely than others, and the neat probabilities in a textbook start to drift away from what happens at the table.
So the practical question is how much shuffling brings a deck close enough to random that the difference does not matter.
The riffle shuffle model
The most-studied method is the riffle shuffle, in which you split the deck into two roughly equal packets and interleave them. Mathematicians describe this with a simple model in which the cut is made at a point drawn from a binomial distribution and the cards are dropped from each packet with probability proportional to the packet’s remaining size. This is known as the Gilbert-Shannon-Reeds model, and it is a mathematical idealisation of how people shuffle, not a perfect description of any individual person.
A single riffle has a visible structure. The result is made of two interleaved runs of the original order, so cards that began in order still appear in order, with others in between. Each additional riffle roughly doubles the number of such interleaved sequences, which gradually breaks the original order apart.
The seven-shuffle result
In the early 1990s, mathematicians Dave Bayer and Persi Diaconis analysed how quickly repeated riffle shuffles approach a uniformly random ordering. They measured the distance from random using a quantity called total variation distance. Their finding for a 52-card deck was that the distance stays close to its maximum for the first few shuffles, then falls quickly, and by about seven shuffles it has dropped to a small value. This is where the popular statement “seven shuffles are enough” comes from.
Some points on interpreting that result:
- It describes a specific measure of closeness. Other measures of randomness can suggest somewhat different numbers, and the result for a particular purpose might be a little lower or higher.
- It uses the idealised model above. Real shuffles are less precise: packets are uneven, cards sometimes clump, and people vary.
- It concerns the deck as a whole. For a particular game you may only care about some features of the order, and those can be mixed sooner than the complete ordering.
- “Enough” does not mean exactly random. It means close enough that distinguishing the deck from a random one would be hard.
Fewer than around four or five riffles leave the deck noticeably non-random under this measure, which is why a quick two or three shuffles at the start of a game is often not very thorough.
Other kinds of shuffle
Different methods mix at very different speeds.
- Overhand shuffling, where you pull small packets from one hand to the other, moves cards slowly. Analysis suggests that a very large number of overhand shuffles is needed to mix a deck, many times more than for riffles.
- Pile shuffling, where cards are dealt into several piles and the piles are stacked, is predictable and does not randomise on its own. It is often used to count or check a deck, not to mix it.
- Cutting does not change the cyclic order of the deck. It moves the starting point but keeps neighbours together, so it should not be counted as shuffling.
- A perfect faro shuffle, which interleaves two exact halves card by card, is completely deterministic. Repeating it eight times with the right type of faro returns a 52-card deck to its original order, which is a good illustration of why precision is not the same as randomness.
Why games make it harder
Card games create a particular problem: the deck does not start in random order. After a game, cards are often collected in groups, such as resource cards together, spells together or cards in the order they were played. Those groups remain largely intact after a few quick shuffles. That is why decks sometimes seem to deal clumps of similar cards.
If you usually put resources together at the end of a game, a thin shuffle may leave them together at the start of the next. The effect is not mystical and does not mean the probabilities are broken; it means the shuffle did not do its job. Ordinary-seeming flood or drought hands may sometimes be partly a mixing issue, though without a careful record you cannot tell that from bad luck, and a feeling of clumping is hard to separate from the usual patterns that occur in random sequences.
Randomness feels streakier than you expect
Truly random sequences contain runs. People tend to expect a more even spread than chance produces, which is a part of why streaks seem suspicious. Seeing three lands in a row does not by itself indicate poor shuffling. Telling the difference requires many games’ worth of observations and a way of testing them, such as a chi-square test comparing observed and expected counts. This ties into the hot hand and gamblers fallacy discussions: the sense that a streak must mean something is strong, and often misleading.
Practical suggestions
- Riffle several times, with the common figure of about seven as a starting point for a full deck, then cut.
- Mix in a different style as well, since combining methods helps break up groups that one method leaves intact.
- Do not rely on cutting or on a single quick shuffle after an organised game ends.
- For decks of 40 or 60 cards, the exact count differs slightly from 52, but the overall idea is similar. Do not treat seven as a precise law.
- In serious competition, follow the shuffling and presenting rules of the event, which typically give the opponent a chance to cut or shuffle.
Randomness in a simulated or digital game comes from random number generators instead, and those have their own questions about quality and seeds.
The main message of the research is encouraging rather than alarming. A modest amount of careful shuffling brings the deck remarkably close to random, but a handful of casual shuffles often falls short of that.