Twice in one year
Steve Gadbois (Mathematics Magazine) and John Emert with Dale Umbach (the journal Chance) independently described the paradox in 1996. NPR radio reported on it the same year.
In poker the stronger hand is the one that is harder to get: without a joker three of a kind turns up less than half as often as two pair, and that is why it beats it. Now the deck has a joker that can become any card, and its holder makes the strongest hand from it. Does three of a kind stay rarer than two pair? Lock in your answer, then check it: deal some cards, count all 2,869,685 hands and read the explanation.
Without a joker, three of a kind turns up less than half as often as two pair. But a pair, two odd cards and a joker can become either three of a kind or two pair, and the holder picks the higher hand. There are 82,368 such hands, all of them go to three of a kind, and with the joker there are 137,280 three-of-a-kinds against 123,552 two pairs. Put two pair higher and those hands go to it instead: now two pair is the commoner one. No order of hands keeps the rule "rarer is stronger".
Poker does without arbitrary choices: nobody invented the order of hands, rarity sets it. Whatever is dealt less often is stronger. Let us see whether this rule survives a single joker.
52 cards, 5 to each player. There are 2,598,960 different hands, and the order of hands follows their rarity exactly: high card is the most common, straight flush the rarest. There are 54,912 three-of-a-kinds and 123,552 two pairs, so three of a kind beats two pair.
Now the deck has 53 cards and 2,869,685 different five-card hands. The joker can become any of the 52 cards, even a copy of one already in the hand. So a new hand appears, five of a kind, and it is the strongest.
Whoever gets the joker turns it into the strongest hand the order allows. For example, the cards 7, 7, king, 4 and a joker can make either three sevens or two pair: sevens and kings.
We keep the usual order of hands: three of a kind ranks above two pair. Which of the two hands turns up more often now?
"Without a joker there are more than twice as many two pairs. One joker in the deck won't turn that round"
In fact it will. A hand with a pair and a joker can become either three of a kind or two pair, and the holder picks three of a kind because it ranks higher. There are 82,368 such hands, and with them three-of-a-kinds reach 137,280 against 123,552 two pairs.
From strongest to weakest, out of 2,598,960 hands.
A 52-card deck plus one joker, 5 cards each, the usual order of hands. The joker's holder always makes the strongest hand. Pick an answer.
A breakdown of your answer will appear here.
Hard to decide? Open the hints one at a time: the easy one first, then stronger ones.
Take a hand with a pair, two odd cards and the joker. Which hands can it become?
Such a hand can become either three of a kind or two pair. The joker's holder takes the one that ranks higher.
Without a joker there are 54,912 three-of-a-kinds and 123,552 two pairs, and 82,368 hands of "a pair, two odd cards and the joker". Which of the two hands do they join?
The next rounds show which hand is more common: deals with a joker, an exact count of every hand and the explanation. Answer round 1 first, or go on?
You chose to always show all rounds.
Cards are dealt from a 53-card deck. When the joker comes, choose what it becomes: the game shows every hand it can make. On the right is the count of hands. Deal thousands of hands at once, then put two pair above three of a kind and count again.
Order of hands:
In the quick deals the joker always becomes the strongest hand, as in a real game.
It seems easy to fix: count the hands with the joker and order them by the new rarity. Try it. The table is an exact count of all 2,869,685 hands for the current order, from the strongest hand to the weakest. The right-hand column checks whether a hand is rarer than the one below it.
The exact numbers above come from going through every hand. Let us check them with random deals: each deal is scored under both orders, and the joker always becomes the strongest hand.
The simulated shares stay near the exact ones, within about 0.02 pp. Under the usual order three of a kind is more common than two pair, and if you swap them, two pair becomes the commoner one.
Once you answer, the simulation code appears here so you can run it yourself.
The same simulation, short and with no libraries. Run the JavaScript in your browser console (F12) or in Node; the Python needs only the standard library.
// A 52-card deck plus one joker. The joker becomes any card, and its holder makes the strongest hand. Which turns up more often: three of a kind or two pair?
const N = 1000000; // five-card deals
const JOKER = 52; // cards 0–51: rank c >> 2, suit c & 3
// 0 high card, 1 pair, 2 two pair, 3 three of a kind, 4 straight, 5 flush, 6 full house, 7 four of a kind, 8 straight flush, 9 five of a kind
function kind(h) {
const n = Array(13).fill(0); h.forEach(c => n[c >> 2]++);
const k = n.filter(x => x).sort((a, b) => b - a);
const flush = h.every(c => (c & 3) === (h[0] & 3));
const r = h.map(c => c >> 2).sort((a, b) => a - b);
const straight = k.length === 5 && (r[4] - r[0] === 4 || r.join() === '0,1,2,3,12');
return k[0] === 5 ? 9 : straight && flush ? 8 : k[0] === 4 ? 7 : k[0] === 3 && k[1] === 2 ? 6 : flush ? 5 : straight ? 4 : k[0] === 3 ? 3 : k[0] === 2 && k[1] === 2 ? 2 : k[0] === 2 ? 1 : 0;
}
const ORDERS = { 'usual order': [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], 'two pair above three of a kind': [0, 1, 3, 2, 4, 5, 6, 7, 8, 9] };
const count = {}; for (const o in ORDERS) count[o] = Array(10).fill(0);
const deck = [...Array(53).keys()];
for (let i = 0; i < N; i++) {
for (let j = 0; j < 5; j++) { const k = j + Math.floor(Math.random() * (53 - j)); [deck[j], deck[k]] = [deck[k], deck[j]]; }
const hand = deck.slice(0, 5), rest = hand.filter(c => c !== JOKER);
// every hand the joker can make (it may copy a card in the hand)
const can = rest.length === 5 ? [kind(rest)] : [...new Set([...Array(52).keys()].map(c => kind([...rest, c])))];
for (const o in ORDERS) count[o][can.reduce((a, b) => ORDERS[o].indexOf(b) > ORDERS[o].indexOf(a) ? b : a)]++;
}
const pct = x => (100 * x / N).toFixed(2) + '%';
for (const o in ORDERS) console.log(o + ': two pair', pct(count[o][2]), '· three of a kind', pct(count[o][3]));
console.log('exact: 4.31% · 4.78% and 7.18% · 1.91%');
# A 52-card deck plus one joker. The joker becomes any card, and its holder makes the strongest hand. Which turns up more often: three of a kind or two pair?
import random
from collections import Counter
N = 1_000_000 # five-card deals
JOKER = 52 # cards 0–51: rank c // 4, suit c % 4
# 0 high card, 1 pair, 2 two pair, 3 three of a kind, 4 straight, 5 flush, 6 full house, 7 four of a kind, 8 straight flush, 9 five of a kind
def kind(h):
k = sorted(Counter(c // 4 for c in h).values(), reverse=True)
flush = len({c % 4 for c in h}) == 1
r = sorted(c // 4 for c in h)
straight = len(k) == 5 and (r[4] - r[0] == 4 or r == [0, 1, 2, 3, 12])
if k[0] == 5: return 9
if straight and flush: return 8
if k[0] == 4: return 7
if k[:2] == [3, 2]: return 6
if flush: return 5
if straight: return 4
if k[0] == 3: return 3
if k[:2] == [2, 2]: return 2
return 1 if k[0] == 2 else 0
ORDERS = {"usual order": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], "two pair above three of a kind": [0, 1, 3, 2, 4, 5, 6, 7, 8, 9]}
count = {o: Counter() for o in ORDERS}
for _ in range(N):
hand = random.sample(range(53), 5)
rest = [c for c in hand if c != JOKER]
# every hand the joker can make (it may copy a card in the hand)
can = {kind(rest)} if len(rest) == 5 else {kind(rest + [c]) for c in range(52)}
for o, order in ORDERS.items():
count[o][max(can, key=order.index)] += 1
for o in ORDERS:
print(o + ": two pair", f"{100 * count[o][2] / N:.2f}%", "· three of a kind", f"{100 * count[o][3] / N:.2f}%")
print("exact: 4.31% · 4.78% and 7.18% · 1.91%")
Without a joker a hand simply is what it is, and its rarity does not depend on the order. With a joker some hands can become several different hands, and the holder picks the higher one. So the order decides how many hands of each kind there are, and how many hands of each kind there are is exactly what we wanted to derive the order from.
“In plain words” is hidden on all puzzles.
A pair of sevens, a king, a four and a joker. Joker as a seven gives three of a kind; joker as a king gives two pair. The holder picks whichever ranks higher. There are 82,368 such "undecided" hands. Without them there are 54,912 three-of-a-kinds and 123,552 two pairs.
Whichever of the two hands you put higher, the undecided hands go to it, and it overtakes the other. This card explains only the pair "three of a kind vs two pair". Why no other order of all the hands helps either is shown in the next layer of explanations.
The rule "rarer is stronger" asks: first count the hands, then order them. But with a joker the count depends on the order. Put three of a kind higher and it becomes more common. Put two pair higher and that becomes more common. It is a closed loop: whichever hand you lift, the lift itself makes it more ordinary.
| Which is higher | Three of a kind | Two pair |
|---|---|---|
| three of a kind | 137,280 | 123,552 |
| two pair | 54,912 | 205,920 |
In red is the hand that ranks higher yet turns up more often. Each time it is the one that was put higher. The loop swings between two orders and never comes to rest.
“Rigorous” is hidden on all puzzles.
Look only at one pair, three of a kind and two pair. An undecided hand (a pair, two odd cards, a joker) can become any of the three: joker as a seven gives three of a kind, joker as a king gives two pair, joker as a deuce leaves just a pair. The holder takes the highest, and the rest of the order does not affect these hands. So everything turns on which of the three ranks highest.
There are no other cases, so in every order some stronger hand turns up more often than a weaker one. The rest of the order can shift only a few thousand hands: 2,808 "two pair and joker" hands can become two pair or a full house, and 2,496 "three of a kind, odd card and joker" hands can become three of a kind, a full house or four of a kind. Hence "at least" and "at most", and they do not change the conclusion. The computer check of all 40,320 orders in round 3 confirms it. The fewest broken pairs, one, occur in exactly two orders: the usual one and the one with three of a kind and two pair swapped.
Three of a kind without a joker: the rank of the three, 3 suits of 4, two other ranks, a suit for each.
Two pair without a joker: the two ranks of the pairs, the suits of each pair, the rank and suit of the fifth card.
Undecided hands: a pair, two cards of other ranks and the joker.
Three of a kind above two pair:
Another 2,496 joker hands (three of a kind, an odd card, the joker) go to four of a kind, not to three of a kind, since four of a kind ranks higher. So the joker hands that reach three of a kind are exactly the 82,368 undecided ones.
There are 270,725 hands with the joker: the joker plus any 4 of the 52 cards. Trying all 52 replacements for the joker splits them into groups by which hands are reachable. The order decides only the fate of the groups that have a choice.
“Deeper” is hidden on all puzzles.
The step "count under the order, then rerank by rarity" turns one order into another. A consistent order is one this step leaves in place: the recount gives the same rarity the order came from. For an ordinary deck such an order exists, and it is the familiar order of poker. With a joker there is none: the step only flips three of a kind and two pair back and forth.
A similar trap sits in Efron's dice and in the Condorcet paradox: a reasonable demand on a ranking that no ranking can meet. There the culprit is a cycle of pairwise wins; here it is a ranking that changes the data it stands on.
Steve Gadbois described the paradox in "Poker with Wild Cards — A Paradox?" (Mathematics Magazine, 1996, vol. 69, pp. 283–285) and counted a deck with two jokers. The same year John Emert and Dale Umbach of Ball State University independently concluded that with wild cards there is no order in which more valuable hands occur less often (Chance, 1996, vol. 9, no. 3). They also noticed that the usual order has the fewest inconsistencies, as round 3 shows too.
Emert and Umbach proposed another rule: count not what a hand became but what it could become. Such an "inclusion" count does not depend on the players' choices, so there is no loop. The price is an order that drifts a little from the usual one, and with more wild cards the flush moves up.
More wild cards mean more broken pairs of hands. And the rules people really play by dodge the paradox in different ways.
With two jokers (54 cards, 3,162,510 hands) there are at least three inconsistencies. Three of a kind: 232,968 against 123,552 two pairs. One pair becomes more common than high card: 1,437,936 against 1,302,540. And full house and four of a kind turn up exactly equally often, 9,360 times each, so "rarer" cannot pick either of them.
If you put one pair below high card, the same undecided hands go to high card, which rises to 1,642,236 against 1,098,240 pairs. The same loop again, now at the bottom of the table.
With deuces wild (52 cards, the four 2s stand for any card) four of a kind turns up 31,552 times, against only 12,672 full houses and 14,472 flushes, though four of a kind beats both. Of the 40,320 orders of the eight lower hands none is consistent, either here or with two jokers, and the usual order is again among the least broken: 3 broken pairs with two jokers, 4 with deuces wild.
The order of hands was only meant to describe rarity. But as soon as it became the rule of the game, the players adapted to it, and the rarity changed. It is the poker version of Goodhart's law: when a measure becomes a target, it stops measuring well what it used to measure. The economist Charles Goodhart put the idea forward in 1975 for monetary policy, and the anthropologist Marilyn Strathern gave it its well-known form.
The general lesson: if a rule rewards something for being rare and people can choose, rarity has to be measured before they start choosing, or on something the choice does not change. That is exactly what Emert and Umbach did: they count what a hand could become, not what it was made into.
Numbers for a deck with one joker; each rule was checked by going through all 2,869,685 hands.
| The rule | What it does | What happens to the paradox |
|---|---|---|
| The usual order, fixed in advance | The order is never recounted: three of a kind is higher because the rules say so. | One broken pair remains: three of a kind is commoner than two pair. No order does better. |
| The "bug", the usual casino rule | The joker may only be an ace or a card that completes a straight, flush or straight flush. | The paradox goes away: the usual order is consistent again (63,360 three-of-a-kinds, 138,600 two pairs), and five of a kind is a single hand, five aces. |
| Emert and Umbach's inclusion order | Rarity is counted by what a hand could become, not what it became. | No loop, because the count does not depend on choices. The order differs a little from the usual one. |
Steve Gadbois (Mathematics Magazine) and John Emert with Dale Umbach (the journal Chance) independently described the paradox in 1996. NPR radio reported on it the same year.
According to the mathematics writer Ivars Peterson, a similar observation about wild cards already appears in "Scarne on Cards" by the gambler and magician John Scarne, first published in 1949.
With one joker, five of a kind comes only from four of a kind plus the joker: 13 hands of 2,869,685. That is almost 16 times rarer than a straight flush (204 hands with the joker).
With two jokers there are exactly as many full houses as four-of-a-kinds: 9,360 each out of 3,162,510 hands. Rarity cannot choose between them at all.
When all the deuces are wild, four of a kind turns up 2.5 times as often as a full house: 31,552 against 12,672 hands of 2,598,960. And four of a kind still ranks higher.
Christopher Williamson (arXiv, 2025) worked out poker for decks with different numbers of ranks. By his account, the order of hands stops changing only from 761 ranks, that is, a 3,044-card deck.
Share your own explanation, an objection or a variant of the puzzle. You need a GitHub account to comment.
Is “School maths” the right level for this puzzle?
Thank you! This helps set the levels right.
I keep adding new puzzles, new ways to explain them and write-ups of the discussions they start. Leave your email and I'll write when something new is out. No spam, one-click unsubscribe.
The sign-up form will appear here soon.