Miraskill
A spectator shuffles the deck however they like and deals it out two cards at a time. Red pairs go on one mat, black pairs on another. A prediction written before the shuffle says how those two piles compare, and it is right.
Watch the lectureLecture 01 · Miraskill: order hidden inside a shuffled deck
Try it
Shuffle the packet and cut it wherever you like, then deal it in pairs and put each pair on its mat. The prediction stays sealed until the last pair is down.
Why it works
Nothing about the order is secretly arranged. The shuffle really decides which cards become partners, and so it decides how many pairs land on each mat. What it cannot decide is how many red cards and how many black cards there are to begin with: twenty-six of each.
Follow the cards rather than the piles. A mixed pair uses up one red card and one black card. Take all the mixed pairs away and the supply is still balanced: whatever was equal before is equal after. The cards that remain on the two outer mats are red cards sitting with red partners and black cards sitting with black partners, in equal numbers. Equal numbers of cards make equal numbers of pairs.
Every mixed pair removes one red and one black, so the cards left for the matching piles are as balanced as the deck you started with.
Below is one twelve-card run from the course, six red and six black, played slowly. Step through it and watch the gold line. It adds two things: twice the lead of red–red pairs over black–black pairs, and red minus black among the cards still in the packet. Every pair leaves that sum at 0, because a matching pair moves 2 from one term to the other and a mixed pair changes neither. When the packet is empty, the second term is 0, so the lead must be 0 too.
Counting it in general
Name the counts, which are numbers of pairs: r is the number of red–red pairs, b the number of black–black pairs, and m the number of mixed pairs. Every red card ends up somewhere: two in each red–red pair and one in each mixed pair. So the number of red cards is 2r + m. In the same way the number of black cards is 2b + m.
With a full deck both totals are 26, so 2r + m = 2b + m. Take away the m on each side and halve: r = b. The shuffle changes r, b and m; it cannot change this identity.
If the deck starts with R red cards and B black cards instead, the same subtraction gives 2(r − b) = R − B. The prediction was never about one card. It was about a relation that every order of the deck must obey.
Challenge: an uneven packet
Openly set aside some cards so the packet is no longer balanced. Before a single pair is dealt, predict r − b: red–red pairs minus black–black pairs. Then try to beat your own prediction by arranging the dealing order yourself.
Perform it yourself
Set-up
- Use one complete 52-card deck with the jokers removed. Count it beforehand: 26 red and 26 black. No stack is needed.
- Write the prediction on a card: THE TWO MATCHING PILES WILL TIE. Put it face down on the table before anyone shuffles.
- Lay out three mats or spaces, labelled RR, MIXED and BB.
Performance
- Hand the deck to the spectator to shuffle in any way they like: cuts, mixing small groups, spreading on the table. Every card stays in the deck.
- The spectator takes the next two cards from the top and turns them over. Two reds go on RR, two blacks on BB, one of each in the middle. Suits and values do not matter: a heart and a diamond are two reds.
- Let them deal every pair. Do not steer the piles; a run of red pairs is welcome. Never omit the last pair.
- Ask them to count pairs on both outer mats, not cards on one and pairs on the other. Lay them side by side, one red pair beside one black pair: the two rows finish together.
- Open the prediction. You predicted a relation between the piles, not the number of pairs in either; that number really does change with the shuffle.
Handling notes
- If a pair has been put on the wrong mat, say so and audit the pairs openly before opening the prediction. Never assert a tie that the table does not show.
- If the spectator predicts that one colour will win, keep that guess and compare it with the finished deal.
- A smaller version shows the whole idea at once: openly take six red and six black cards and deal six pairs.
- Stopping halfway is a different experiment. A pile can lead early; the promise concerns the completed deal.
Going deeper
Why an early lead does not matter
Suppose a twelve-card packet (six red, six black) opens with two red–red pairs. The red pile leads by two pairs, and four red cards are gone. The packet that is left holds two red cards and six black cards. Pair those eight in any way: 2(r − b) = 2 − 6, so they must produce two more black–black pairs than red–red pairs. That cancels the early lead exactly.
Nothing says the next pair must be black–black; it might be mixed, or the last red–red pair. What is forced is the total once every card is used. A global guarantee need not prescribe the next local event, and no probability is involved: a deliberately arranged packet obeys the same count.
The hypotheses, stated carefully
The tie needs two things: equal numbers of red and black cards at the start, and every card used in a pair. Remove four red cards and the packet has 22 red and 26 black; now the black–black mat must win by two pairs, which is four cards. Say which scoreboard you mean.
With an odd packet, one card is left without a partner. That card's colour belongs in the count; you cannot drop it and use the old equation unchanged.
The magician's question: what cannot change?
A shuffle can change which card is on top, but it cannot change how many red cards there are. An invariant is a quantity like that, and it is only meaningful together with a list of allowed moves: shuffling preserves the red count, but swapping a red card for a black one would not. Proving an invariant takes two steps: check that it holds at the start, then check that each allowed move keeps it.
In Miraskill the invariant can be watched during the deal. Count red minus black among the cards already on the mats, and red minus black among the cards still in the packet. Each part changes as pairs are dealt; their sum stays at its starting value, zero. A single legal counterexample is enough to refute a claimed invariant: “the top card stays red” fails for red, black after a one-card cut.
Sources
The principle is Stewart James's Miraskill. The account here follows Persi Diaconis and Ron Graham, Magical Mathematics (Princeton, 2012), Chapter 10, p. 184; the presentation and the twelve-card teaching replay are adapted for this course.