Spades — with the rules the free sites leave out

The spades card game, with the two rules the free sites drop

Spades is an American trick-taking game from the late 1930s — devised in the Midwest, and traced by the games writer George Coffin to Cincinnati between 1937 and 1939. It is played by four people in two partnerships, with an ordinary 52-card pack and no jokers, and it is entirely in the public domain. It became the card game of American college dorms, military barracks and, later, Windows, which is why so many people know the shape of it and so few have played a version that scores it properly.

Because two rules decide this game and they are the two that get cut. The first is bags: every trick your side takes beyond its bid is worth one point now and costs you a hundred later, and the counter runs across hands rather than resetting. The second is nil — and specifically what happens when a nil fails. Both are implemented here, both have named tests in our suite, and the bag counter is on the board at all times rather than buried in a scorecard you have to open.

How to play spades

Partners sit opposite each other, so the turn order alternates between the two sides. The whole pack is dealt out one card at a time, thirteen each, starting to the dealer's left.

The bid

Starting to the dealer's left and going clockwise, each player says how many of the thirteen tricks they expect to take. The two partners' bids are then added together into a single contract. This is the part beginners are taught wrongly: your own number does not matter on its own. If you bid 4 and your partner bids 2, your side has to take at least six tricks between you, and it does not matter which of you takes them.

The play

The player to the dealer's left leads to the first trick. Everyone must follow suit if they can. If you cannot follow, you may play anything — including a spade, which wins the trick. Highest card of the led suit takes it unless a spade was played, in which case the highest spade takes it. The winner of a trick leads the next one. The ace is high.

When you may lead a spade

You may not lead a spade until spades are broken, and spades are broken the first time somebody plays one on a trick that was led with a different suit — normally because they were void and trumped in. There is one exception, and it is the half that clones drop: if spades are all you have left, you may lead one anyway. Without that exception a player down to nothing but spades has no legal move at all, and we have played implementations where exactly that happens.

On this table you never have to remember any of it. On your turn, the cards you may legally play are drawn on a bright cream face and the ones you may not are greyed out, live, before you touch anything. If you tap a grey card the board tells you which rule stopped you rather than ignoring the click.

Bags — the rule that makes spades a game

Take more tricks than your side bid and each extra trick is a bag. A bag is worth one point immediately, which looks like a bonus. It is not. Bags accumulate across hands, and the tenth one costs your side 100 points. Anything past ten rolls over into the next block of ten — nine bags plus three more is twelve, which is one 100-point penalty with two still on the counter.

The consequence is the whole strategic core of the game: once your side has made its contract, you want to lose the rest of the tricks. Every trick you win from that point is a bag on the counter, and about one in every eight of them costs a hundred points. Skilled spades is largely the art of deliberately playing a losing card, and a version without bags is a different, much shallower game that happens to use the same deck.

This is not a variant we chose to add. Pagat gives it as the standard rule: "A side which (over several deals) accumulates ten or more bags has 100 points deducted from its score. Any bags beyond ten are carried over to the next cycle of ten overtricks." The reason we labour the point is that most free browser spades either omits bags entirely or shows them without ever applying the penalty, and both produce a game where the correct play is simply "win everything you can", which is the opposite of spades.

Nil, blind nil, and the detail nobody implements

Nil is a bid of zero tricks. Make it — take nothing at all — and your side scores 100. Take a single trick and your side loses 100. Meanwhile your partner still bids normally and their contract is scored exactly as usual: the nil is a separate bet alongside it, not a replacement for it. A side with a nil bidder and a partner who bid 4 makes 140 points for a clean nil and a made contract.

And here is the detail that almost no implementation gets right. When a nil fails, Pagat is explicit: "the tricks won by the nil bidder do not count towards making the partner's bid, but do count as bags for the team." So a busted nil is worse than it looks — it costs 100, it feeds the bag counter, and it cannot rescue a partner who is a trick short of contract. Our engine implements exactly that asymmetry and has a named test for the case where the busted nil's tricks would have made the partner's bid and are not allowed to.

Blind nil is a nil declared before you have seen a single card, and it pays double: 200 for a clean one, minus 200 for a busted one. It is offered here only to a side trailing by 100 points or more. On this table it is real rather than cosmetic: while the option is open your hand is genuinely dealt face down, and pressing Look at my hand forfeits the blind. That is enforced in the game engine — the engine refuses a blind bid from a seat that has looked — rather than being a thing the interface politely hides.

Some houses have the blind-nil bidder swap two cards with their partner. That is not implemented here and the page says so rather than letting you find out at the worst moment.

Scoring, in full

The default target is 500, which Wikipedia gives as the usual condition; 300 and 200 are on the start screen for a shorter sitting. If both sides cross the target in the same hand the higher score wins, and an exact tie plays another hand.

One place the published rules are genuinely silent, so we made it a setting rather than an assumption. Pagat says a failed nil bidder's tricks "count as bags for the team" but does not say what happens when that side is also set. Every implementation we could check gives a set side no bags at all, so that is our default — but both behaviours are in the engine and both have tests, and this paragraph exists so nobody has to guess which one they are playing.

A spades deal you can actually check

Every online card game 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.

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 the bag counters, not whether you are one hand from 500. 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 all four thirteen-card hands. 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 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 of the array, one card at a time, starting at the seat to the dealer's left and going clockwise for thirteen rounds; the pack is exhausted exactly.

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 the deal, not that a modified opponent client could not look at your thirteen cards. Hiding a hand from the computer that dealt it is a different problem — mental poker — and we have not built it. 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, so you can satisfy yourself the hash is really SHA-256 before you trust anything computed with it.

The arithmetic underneath, and where it is checked

A spades hand is thirteen cards from a 52-card pack, which makes it the same object as a bridge hand — and bridge deal combinatorics are the most thoroughly published in all of card games. So unlike most of the games on this site, this one has a real external anchor to check the shuffle against, and we reached it.

There are 635,013,559,600 distinct thirteen-card hands. Our suite recomputes that figure three independent ways in exact integer arithmetic — a falling product, Pascal's recurrence, and the raw factorial ratio — and asserts all three against the published number rather than quoting it. It then derives the whole hand-pattern table from first principles and matches the published percentages to five decimal places:

Then the part that matters for a card game: the same numbers are asserted against our own dealer. The number of spades in a thirteen-card hand follows the hypergeometric distribution — P(k) = C(13,k)·C(39,13−k)/C(52,13) — which the suite derives exactly, proving along the way that the mean is 13 × 13/52 = 3.25 spades exactly and the variance is exactly 507/272, both as integer identities rather than as floats that could launder a rounding error into a pass. It then deals 200,000 real hands out of the real shuffler on a fixed seed and compares: chi-square 8.66 on 13 degrees of freedom against a 99.9% critical value of 34.53, no single bucket more than 4 sigma out, and the observed mean 3.25 within 4 sigma of exact. The five commonest published patterns come out of our dealer within 4 sigma too, which closes the loop from a published number, through our own derivation, to our own cards.

Two smaller figures fall out of the same machinery and are asserted exactly: the chance that one player is dealt all four cards of a named rank is exactly 143/54145 — 0.00264, or one deal in 379 — and the chance of a hand with no spades at all is C(39,13)/C(52,13) = 0.01279, about one in 78.

The honest ceiling. Walking all 635,013,559,600 hands is not something this suite does, or could. What is exact is the closed-form work — the binomial coefficients, the pattern counts, the hypergeometric distribution and its moments. What is sampled is the dealer's output, and its sampling error is stated in the suite rather than implied away.

What is NOT anchored, and why we are saying so

There is no published table of expected tricks by hand strength for spades. We looked: Wikipedia's spades article contains no bidding table of any kind, and Pagat's page is a rule book with no numbers in it. Bridge's fifty years of hand-evaluation literature does not transfer, because bridge has a variable trump suit, a contract auction and a dummy. So the bidding model in our bot is ours, and every figure below is our measurement against our own engine, not a published result:

• Over 10,000 seats where every player was made to bid 3 so that nils could not distort the sample, our trick estimator averaged 3.18 against an actual mean of exactly 3.25, with a mean absolute error of 0.75 tricks and a correlation of 0.78. A trick estimator in a hidden-information game cannot be accurate; it can be unbiased, which is what makes a bid a fair contract rather than a systematic overbid.
• At a real table where nils are allowed, the same estimator reads about 0.4 tricks low for every player who did not bid nil — not because the model got worse but because roughly a sixth of seats bid zero and give their share of the thirteen tricks to everyone else.

What each bot level actually does

None of them search ahead — spades gives a player three decisions (what to bid, what to lead, what to follow with) and each level differs in what it knows, not in how deep it looks. Every win rate below was measured over games with the seats swapped, so a dealer or lead advantage cannot be read as strength.

Easy counts its aces, its high spades and its spade length, bids that, and then plays a legal card at random. It never bids nil and it has no idea what a bag is.

Medium wins a trick whenever it can do so with its cheapest winner and otherwise throws its lowest card, and it bids nil on hands that look safe. No partner model, no memory, no bag sense — which is, as far as we can tell from playing them, exactly the bot every free spades site ships. It beats Easy 96% of the time.

Hard adds the three things that actually separate a spades player from a card player. It tracks who is void in what, because every time somebody fails to follow suit the whole table learns something permanent. It has a partner model: it does not overtake its own partner's winning card, it takes the trick off a partner who bid nil rather than letting the nil bust, and it will not lead a suit its nil partner is void in — which is the classic tell of a bad spades bot, because that lead forces the partner to ruff and win. And it counts bags: once the contract is made it stops taking tricks. It beats Medium 61% of the time and takes measurably fewer bags per hand (0.92 against 1.09).

The honest limits on those numbers, and one finding we left in rather than tuned away. 61% means Hard beats our Medium; it is not a claim that Hard is a strong spades player, and it has never played a human expert or any published agent. And the levels are not a total order: Medium beats Easy 96% while Hard beats Easy only 92%, reproducibly. The cause is nil — a random opponent busts nils by accident that a competent opponent would let through, so the very aggression that makes Hard beat Medium costs it against Easy. We swept the nil threshold across six settings and chose one with both columns on the table instead of tuning to whichever sparring partner was handy; the full sweep is in the engine source. A policy tuned against a single opponent is not evidence of general strength.

What this table does and does not do

It plays standard four-handed partnership spades to 500 (or 300, or 200). Here is everything deliberately left out, so you do not have to discover it:

• Live play seats two people, not four. You and your opponent sit opposite sides of the table and each of you has a house bot as a partner. Four humans in one room needs four-way seating, a roster and three separate ways for a seat to vanish mid-hand, and it is not built. This is the biggest limitation on the page and it is why it is first.
• No card exchange after a blind nil. Some houses have the blind bidder swap two cards with their partner; we do not.
• No jokers, no deuce-of-diamonds, no "10 for 200", no mirror, suicide or whiz bidding, no three-handed or cutthroat spades.
• No minimum team bid. Some houses require a combined bid of at least four; this table does not enforce one.

And a phone is tight, measured rather than glossed. You hold thirteen cards from the first trick to the last and they all have to be on screen while you decide, which at 390px is about a 28-pixel card — well under the ~44px a fingertip wants. There is no layout that fixes that; thirteen cards is thirteen cards. What we do instead: the tappable area of each card claims the gaps either side and reaches above and below it, and a tap only picks a card up — nothing is played until you press the button that names it. So a mis-tap costs you one more tap and can never cost you the wrong card.

Questions people ask

How do you play the spades card game?

Four players in two partnerships, thirteen cards each. Everyone bids the tricks they will take, partners' bids are added into one contract, and then you play thirteen tricks following suit where you can. Spades are always trump. Make your contract and score ten times the bid; miss it and lose ten times the bid. First side to 500 wins.

When can you lead spades?

Only after spades are broken — that is, after somebody has played a spade on a trick led with a different suit. The exception is that you may lead a spade if spades are all you have left.

What are bags?

Tricks over your side's bid. Each is worth one point now, they accumulate across hands, and every tenth one costs your side 100 points with the remainder carried over.

What is nil?

A bid of zero. Take no tricks and your side gets 100; take any and it loses 100. Your partner's bid is scored exactly as normal alongside it. If the nil fails, the nil bidder's tricks do not help the partner's contract but do become bags.

What is blind nil?

A nil declared before you look at your cards, worth ±200, offered only when your side is at least 100 points behind. On this table the hand really is face down until you choose.

How many spades hands are there?

635,013,559,600 — the same as a bridge hand. The average hand contains exactly 3.25 spades, and about one hand in 78 has no spades at all.

Can I play spades with a friend?

Yes — Private room gives you a four-letter code and a link. Note that live play seats two humans as opponents, each with a bot partner; four humans in one room is not built.

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 spades free?

Yes. Three bot levels, live play, every hand, no account, no coins, no chips, no buy-in, no daily limit, no adverts between you and the table. Nothing here is locked and nothing is sold.