In the card game Dobble (Spot It!), any two cards you ever lay down share exactly one picture: never zero, never two. That is not careful design or luck. It is a theorem: the retail deck comes from a finite projective plane with two cards removed. Play the game below; the deck is built in front of you from nothing but arithmetic.
Here is the whole trick of the game, stated once. Think of each card as a line and each symbol as a point. One of the axioms of a projective plane says:
any two lines meet in exactly one point.
Translate it back: any two cards share exactly one symbol. The game's only law and one of the plane's axioms are the same sentence. So if we can build a projective plane, we get a perfect Dobble deck for free, and we can, out of arithmetic over a finite field.
Two cards. They share exactly one symbol. Find it — click it on either card — as fast as you can. New pair, again. Forever. Watch: you will never be dealt a pair with no match, and never one with two.
That last counter is the honest part: it ticks up only if the live deck ever deals a pair sharing zero or two symbols. It stays at 0 no matter how long you play, because the construction makes it impossible — not because we filtered the bad pairs out.
Pick a prime 3. Every symbol is a direction in a tiny 3-D space over the numbers {0,1,…,q−1} with arithmetic done mod 3 — a triple [a,b,c], where scaling the whole triple by a nonzero number gives the same symbol (these are homogeneous coordinates). Every card is also such a triple [L], read as a recipe. A symbol [x] sits on a card [L] exactly when
(L₀·x₀ + L₁·x₁ + L₂·x₂) mod q = 0.
That dot-product being zero is the entire incidence rule. Now the magic: two different cards [L] and [M] share the symbols [x] with both L·x=0 and M·x=0 — and two independent linear equations in this projective space have exactly one common solution direction. Exactly one shared symbol, forced by linear algebra. Swap the roles of point and line and the same sentence holds — the plane is self-dual, so any two symbols also share exactly one card.
Below: the shared-symbol count for every pair of the first up to ten cards. Every off-diagonal cell is 1. No cell is 0 or 2.
For a plane of order n the arithmetic forces the counts exactly: n²+n+1 symbols, the same number of cards, and n+1 symbols on each card. Retail Dobble shows 8 symbols per card, so its order is 7: a full deck of 7²+7+1 = 57 cards drawn from 57 symbols. The box ships 55 — two cards dropped, a manufacturing choice, not a mathematical one (removing cards can never create a second match).
So which orders n can exist at all? This is where the truth gets strange and we say exactly what is known:
| order n | plane exists? | why |
|---|---|---|
| 2,3,4,5,7,8,9,11,13… | yes | every prime power: the same construction, using arithmetic in the finite field of that order |
| 6 | no | Bruck–Ryser (1949): 6 ≡ 2 (mod 4) and 6 is not a sum of two squares ⇒ impossible |
| 10 | no | Bruck–Ryser is silent (10 = 1²+3²); eliminated only by exhaustive computer search — Lam, Thiel & Swiercz, 1989, thousands of CPU-hours |
| 12 | OPEN | nobody on Earth knows. Not built, not forbidden. The smallest undecided order. |
The "found" counter is named for the human eye; "pairs dealt with ≠1 shared symbol", for the limit of the proven. A children's matching game sits on top of a question (does a projective plane of order 12 exist?) that has been open since before the game was invented.
The check. The browser builds the selected deck, counts its cards, checks each dealt pair and the displayed matrix, and runs the Bruck–Ryser test. The retail count and historical dates are sourced; the 51/51 result in research/dobble-projective-plane/verify.mjs records the offline checks: the counts for orders 2,3,5,7; that every pair of cards shares exactly one symbol over all C(n²+n+1, 2) pairs; the dual statement for symbols; the retail order-7 reconciliation; and the Bruck–Ryser obstruction for order 6 (and its silence on order 10).
Honesty notes. The page constructs the planes of prime order and checks the matching property on what it builds. The nonexistence of order 10 is cited (Lam–Thiel–Swiercz 1989), not re-run — it took a supercomputer. Order 12 is genuinely open; we make no claim about it. The symbols are little glyphs drawn for play; the mathematics is in the incidence, not the pictures.
Projective plane construction: PG(2,q) over the field F_q. The Dobble↔projective-plane identification is folklore in recreational mathematics; the existence theory is Bruck–Ryser (1949) and Lam–Thiel–Swiercz (1989). Spot It!/Dobble is a trademark of its publisher; this page uses no game assets.
Written 2026-06-21. Claims re-read against their sources on 2026-10-14: 159 checked, 124 confirmed, 34 wrong, 0 unverifiable, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/cited-1014b/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent.
This pass implements the directing triage for every-two-cards at 67088387e0, after the Codex fixer independently reread the evidence for all 35 triage items. It records those items separately, carries forward each remaining checker-confirmed source entry once with its checker attribution, and records one additional MINOR verifier-label correction. The triage takes precedence over earlier checker confirmations of the same wording. The wrong count includes MINOR claims; the sole STALE item is the JSON-LD date. Literal \u2014 and \n encode punctuation and line breaks in original claim quotations.
This record is long (159 claim lines and 35 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.