Crazy Eights — with the suit in play always on screen

Play crazy eights with a friend, or with real people

Press Private room and you get a four-character code and a link. Send either one to whoever you want to play and they are sitting opposite you — no account, no app, no email address on either side, and nothing to install. If you would rather play a stranger, Find an opponent drops you into the oldest crazy eights room that is waiting; if nobody is waiting it opens one for you and seats the next person who goes looking. Live games count toward a rating; bot games do not, because a browser can lie about its own score and we would rather label that than launder it.

A dropped connection does not end the game. Refresh the page, close the laptop, lose signal in a tunnel — the same link puts you back in your own seat with the hand, the score and the history exactly where you left them. While you are away the house bot covers your turns so your opponent is not sitting in front of a dead table, and it hands the seat straight back the moment you return.

Live play is two-handed. Three- and four-player tables run against bots. Seating three humans in one room is a real feature and it is not built yet; the page says so rather than letting you find out after you have sent the link.

The rule everybody gets wrong, and why it is on the table instead of in the rules

An eight is wild. You may play it on anything, and when you do you name the suit the next player has to follow. The suit you named — not the suit printed on the eight — is what is in play from that moment. Play the 8♠ and call hearts and the next player must produce a heart or another eight; a spade will not do, even though there is a spade sitting face up on the pile.

That single sentence is where implementations of this game go wrong, and the reason is mechanical: the obvious way to write the rule is "does this card match the suit of the top card", which is correct for every card in the pack except the one the game is named after. Our engine keeps the active suit as a separate piece of state and never reads the top card's suit to decide legality — there is a test that plays the 8♠, calls hearts, and asserts both that a heart is accepted and that a spade is refused. There is a second test for the place the same bug hides twice: when the stock runs out and the discard pile is reshuffled, the named suit has to survive the reshuffle.

So the suit in play is drawn on the table, large, at all times, and it says plainly when it is there because somebody called it. This is the one thing we would keep if we had to throw the rest of the page away.

Four rule books, four different draw rules — so you pick

Nobody agrees how many cards you draw when you cannot play, and almost no site tells you which version it has implemented. We read the published rules of the four largest implementations and the two most-cited rule books on 1 August 2026, and got four different answers:

All four are on the start screen and the default is "until you can play", which is what Pagat and Bicycle both say. They are not cosmetic variations: we simulated them and a two-handed hand runs 32.6 turns under draw-until-playable against 22.6 under draw-one, and the stock has to be reshuffled in 23% of hands against 0.8%. Those are our measurements, over thousands of games, and the confidence intervals are printed by the test suite.

There is a fifth disagreement, and it goes the other way. May you draw when you could play? Pagat says yes and treats it as the normal game — "you may always use your turn to draw a card. However, some people play that you may only draw if you are unable to play." cardgames.io forbids it outright: "You can't draw if you have any cards that you can play in your hand." A reviewer of one of the big mobile crazy eights apps asks for Pagat's version by name: "Just let me draw if I feel like it. Without that option it loses the ability to strategize." Our default is Pagat's, so the Draw button stays live when you are holding a playable card — sometimes taking a card is better than being forced to break up a suit you were saving.

A crazy eights shuffle you can actually check

Every card game online gets the same accusation and gives the same non-answer. Players say the deal turns against them; operators reply that their random number generator is secure. Both can be true at once, which is exactly the problem — a secure generator says nothing about the code that decides when to call it. The complaint is unfalsifiable from your side of the screen, and that is what makes it corrosive. It is the single loudest complaint about the free web version of this game: our notes in bar/crazyeights/LOG.md quote a dozen of them, dated, and the incumbent's own answer is a page of prose beginning "No. The cards are always randomly shuffled." That is a reassurance, not a proof.

So this table does not ask you to believe anything. Before a single card is dealt it publishes SHA-256 of a secret seed — the commitment — in the panel beside the board, and you supply a seed of your own. Every shuffle is a pure function of those two values and the hand number, and of nothing else: not the score, not who is winning, not whether you are one hand from the game. When a hand has been scored we hand you that hand's seed. When the game ends we hand you the master seed. You hash it, check it matches the commitment you were shown at the start, and re-derive the exact 52-card order and the exact cards you were dealt. Press Check the deals and the page does it in front of you; press Copy the receipt and you get every input, every output and the derivation in a dozen lines, so you can redo it in any language you like without our code in the room.

The derivation is deliberately small enough to re-implement: hash master : n to get the hand's seed, hash handSeed : yourSeed : n : counter for a stream of bytes, then run an ordinary Fisher–Yates over the numbers 0 to 51 taking each swap index straight from the next byte — rejecting the few values at the top of the byte range that would make low cards very slightly likelier, because 256 is not a multiple of 52 and that skew is precisely the "predictability somewhere" a suspicious player would be right to look for. A card is rank × 4 + suit. The deal comes off the top: with n players the seat left of the dealer takes deck[51], the next deck[50], and so on round the table for as many rounds as there are cards each; then the upcard, then the stock downward.

And here is what it does not prove, because a hash that reassures without proving anything is worse than no hash at all. Against the bots the whole thing runs in your own tab — there is no server, so this proves the deal was fixed before the first card and matches the seed, and it proves nothing about a machine somewhere else. In a live game one seat derives the deck, which means that browser is holding the seed while the hand is played: the protocol proves nobody chose your cards, not that a modified opponent client could not look at them. Hiding a hand from the computer that dealt it is a different problem — mental poker — and we have not built it. One more limit specific to this game: when the stock runs out the discard pile is reshuffled, and that reshuffle comes off the same committed stream, so it was fixed before the first card too — but re-deriving it by hand needs the whole play history, which the receipt does not carry. The initial deal is what the receipt proves. The same protocol on our dice has a SHA-256 self-test against the published FIPS 180-4 vectors; it is the same hash function this page uses.

How to play crazy eights

Crazy eights is a shedding game for two to seven players with an ordinary 52-card pack. It first appears as Eights in the 1930s (David Parlett, The Oxford Dictionary of Card Games, 1996) and the name Crazy Eights dates from the 1940s. It is public domain.

Each player is dealt five cards — or seven if there are only two of you. One more goes face up to start the discard pile and the rest becomes the stock.

On your turn you play one card that matches the top of the pile, by suit or by rank. The 7♠ can be followed by any spade, or by any other seven. If you cannot play, you draw (see the four rules above), and if you still cannot play you pass. Any eight may be played at any time on anything, and the player of an eight names the suit that the next player must follow.

The first player to get rid of every card goes out and wins the hand. Everybody else's remaining cards are counted, and the winner banks the total. First to 100 points wins the game.

When the stock runs out — and the one place the sources flatly contradict each other

Pagat and Wikipedia say the discard pile is turned over and shuffled to make a new stock, except for the top card, which is still the card in play and stays where it is. Bicycle says the opposite — "If unable to play when the stock is exhausted, the player must pass", with no reshuffle at all. This is not a subtle difference of emphasis between the two most-cited rule books; it is a different endgame. We implement the reshuffle, and the no-reshuffle version is available.

If nobody can move and there is nothing left to draw, the hand is blocked: nobody went out, and the lowest card count wins the hand, scoring the difference against each other hand. An exact tie for the lowest scores nobody — we would rather say that than invent a tiebreak the rule books do not have.

⚠️ A finding we did not expect, and it changed a rule. Turning the pile over an unlimited number of times makes this game fail to terminate, and not rarely: in our first simulation, 5 of 800 four-handed games ran past six thousand turns and were still going. The cycle is exact — once nearly every ordinary card has been drawn into somebody's hand, the discard pile is essentially just the four eights, so a stuck player recycles them into the stock, draws until they hit one, plays it, and the next player does the same thing forever. No hand ever shrinks, and nobody can pass, because an eight is always playable. So the pile is turned back once by default; when the stock runs out a second time it stays empty and the hand blocks normally. That makes the game provably finite rather than hopefully finite, and it is why our blocked-hand rate is higher than the one published figure we could find.

Scoring, in full

Every published source we checked agrees on those numbers, and none of them gives an ace as 11 or an eight as anything other than 50. That is worth saying because it is the one part of this game nobody argues about, and because "eights are 50" is what makes holding one an actual decision rather than a free option: an eight left in your hand costs more than ten ordinary cards.

Which direction the points run is argued about. Dice Game Depot calls "the winner of each hand earns the points left in all other players' hands" a named variant; the other convention is that each loser banks their own hand as penalty points and the lowest total wins. We play the first, and the second is implemented in the engine.

The arithmetic underneath, and where it is checked

There is no published solve of crazy eights. No perft table, no optimal-play result, no published figure for how long a game lasts. We looked: arXiv full-text search for "crazy eights" returns nothing, and the closest published complexity result is about generalised UNO — Demaine, Demaine, Harvey, Uehara, Uno and Uno, The complexity of UNO (arXiv:1003.2851; Theoretical Computer Science 521, 2013), which proves a single-player version NP-complete for an unbounded number of colours and numbers, and shows the fixed-size case is in P. That result does not say crazy eights is hard, and we are not going to pretend it does.

What is exactly computable is the deal, and one number in particular decides how every game opens: whether you were dealt an eight. An eight is the only card that is always playable, so it is the single biggest fact about a starting hand.

So in a two-handed game — seven cards each — you start with a wild card slightly less than half the time, and at a four-handed table, five cards each, it is about one deal in three. Both figures are derived by complement in exact integer arithmetic and then cross-checked four independent ways that share no code path: the multiplicative binomial formula, Pascal's triangle (addition only, so it cannot share a division mistake), factorials, and the hypergeometric sum that counts hands containing an eight directly instead of by complement. Then the suite enumerates all 2,598,960 five-card hands one at a time and counts — nothing sampled — and finally runs a Monte Carlo over the engine's own deal, because a formula can be right about a deck the software does not actually deal.

What we measured ourselves, labelled as ours

Nobody has published mean game length or stock-exhaustion figures for this game, so these are our simulations of our rules settings, with 95% confidence intervals, and the test suite prints them on every run.

The one published quantitative figure we found, cited and compared. Mark Newheiser simulated five million games per strategy matchup and published stalemate rates — "about 1 in every 400 games for 4 player games, and once every 27,000 games for 2 player games" (markmywords.substack.com, 3 July 2021). Our blocked-hand rate is much higher: 4.4% at two players against his 0.004%. It is not a like-for-like comparison and we would rather say so than quietly claim agreement — he does not publish his draw rule or his recycle policy, and both of those dominate this statistic. Our own number moves from 4.4% to 0.0% on the same engine just by switching the draw rule, and our recycle is bounded at one for the termination reason above, which mechanically produces more blocked hands than an unbounded one.

Strategy, briefly

Save your eights, but not to the death

An eight is the only card that can never leave you stuck, so spending one early to avoid a single draw is a bad trade. We tested four different "spend it earlier" rules in the bot — when an opponent is close to going out, after a set number of turns, when the stock is low — and every one of them lost. But an eight left in your hand at the count is 50 points, more than ten ordinary cards, so holding one into a hand you are clearly losing is the most expensive mistake in the game.

Get rid of your tens and picture cards while the hand is young

They are 10 points each and they are the cards you will still be holding when somebody else goes out. Our Medium bot does nothing except shed its most expensive legal card and hold its eights, and that alone beats a bot that plays a legal card at random 71% of the time.

Watch what your opponent could not answer

When somebody draws, you learn something exact: at that moment they held no card of the suit in play, no card of the top card's rank, and no eight. That is the only information the game gives you, and it is what tells you which suit to call when you finally play your wild.

What this table does and does not do

It plays standard crazy eights for two to four players. There are no action cards — no skips, no reverses, no draw-twos, no wild-draw-fours. Those belong to the branded 1971 derivative rather than to crazy eights, and adding them would be copying the one part of that game which is actually owned. Partnership play, Crazy Eights Countdown and Mau-Mau's spoken penalties are also not implemented.

Live play is two-handed only, and three- or four-player tables are against bots. There is no "one card left" announcement — neither we nor the incumbent have one, and in a game with no penalty for failing to announce it, it would be decoration.

What each bot level actually does — and the honest limit

None of them search. Crazy eights asks a player for two decisions — which legal card to play, and which suit to name with a wild — and that is all each level decides. Every win rate below was measured over hundreds of seeded games with the bot placed in every seat in turn, so a seat advantage cannot read as strength. The measuring rig checks itself first: a bot played against a copy of itself returns exactly 50.0% at two players and exactly 25.0% at four.

Easy plays a legal card at random, eights included. It is not a crippled Hard; it is a beginner, and it is why it is still holding a 3 at the count while its wild card went on turn two.

Medium sheds its most expensive legal card and keeps its eights until it has nothing else legal. It beats Easy 71.1% of the time.

Hard adds one ply of real lookahead — of the cards still in my hand, how many would be legal after this one goes down — and a memory of which suits each opponent has failed to answer, used when naming a suit with a wild.

🔴 And here is the part a normal games site would not print. At two players, Hard is not better than Medium. It measures 50.2% against it — a coin flip. We tried nine different variations and every single one landed between 49.0% and 51.7%. At four players the same lookahead is worth a real ~2.8 percentage points, because three cards land between your turns and keeping an answer matters. The reason is measurable and we measured it: in a two-handed game the average turn offers 2.15 legal cards, 39.5% of turns offer exactly one, and on another 27.3% you have none and must draw. Two-handed crazy eights barely has a decision to make, so no bot can be much better than any other, and we would rather publish that than sell you a difficulty ladder that is mostly theatre.

One more measured warning, because it points the other way to intuition: the opponent-suit model loses if you let it decide anything. Weighted enough to overrule what the bot actually holds, Hard drops to 43.9% — six points worse than having no model at all. It is now weighted so that it can only ever break a tie between two suits the bot holds equally, and never overrule the count. That is a guarantee rather than an observation: the model's score is capped at 1 and multiplied by 0.1, which is strictly less than the smallest possible gap between two suit counts.

Crazy eights, and the game it became

If you have played UNO you already know how this works, and the connection is not a coincidence — it is documented. UNO was invented in 1971 by Merle Robbins, a barber in Reading, Ohio, to settle an argument with his son about the rules of crazy eights. He and his family mortgaged their home to fund the first five thousand decks and he sold them from the barber shop; the rights went to International Games for $50,000 plus a royalty of ten cents a game, and Mattel acquired the company in 1992. The source for all of that is his obituary in the New York Times, 16 January 1984.

Crazy eights is the older game and it is public domain — the shedding mechanic, the wild card and the suit call belong to nobody. What is owned is a brand and a specific set of card designs, and neither appears anywhere on this page or in our code. The family is large: Pagat lists Switch in England, Mau-Mau in Germany, Pesten in the Netherlands, Prší in the Czech Republic, Makao in Poland and Tschau Sepp in Switzerland, and the commercial WHOT! predates the 1971 game by decades.

Questions people ask

How do you play crazy eights?

Five cards each, or seven with two players. Match the top of the discard pile by suit or rank. If you cannot, draw. Any eight is wild and its player names the next suit. First to shed every card wins the hand and banks the value of everybody else's cards. First to 100 wins the game.

What does an eight do?

It is wild — playable on anything — and you name the suit the next player must follow. That named suit, not the eight's printed suit, is what is in play.

How many cards do you deal?

Five each, or seven in a two-player game. Bicycle's rules say a flat five; Pagat and Wikipedia give the two-player exception.

How many cards do you draw if you cannot play?

The sources disagree four ways: unlimited (Pagat, Bicycle), three (cardgames.io), one (worldofcardgames, gamerules.com) or five (cardzmania). All four are switchable here.

What happens when the draw pile runs out?

The pile is turned over and reshuffled except the top card. If nobody can move and there is nothing to draw, the hand is blocked and the lowest count wins it.

How do you score crazy eights?

Eights 50, K/Q/J/10 ten each, ace 1, everything else face value. The whole pack is 508.

Is crazy eights the same as UNO?

No — it is the older, public-domain game that UNO was invented in 1971 to settle an argument about. Crazy eights uses an ordinary pack and has only one special card, the eight.

Is the deal rigged?

Not here, and you can check rather than believe. The hash of a secret seed is published before the deal, you add a seed of your own, the seed is handed over afterwards, and the exact shuffle re-derives from it. What it does not prove is set out in full above.

Is this crazy eights free?

Yes. Three bot levels, two to four players, live play against people, every hand, no account, no coins, no chips, no daily limit, no adverts between you and the table.