Codewords

The crossword that fired its clue writer

Take a finished crossword — every square filled, every word interlocking — and delete the clues. Then delete the letters too, and write a number in each square instead: 1 wherever an A stood, 2 wherever a B stood, and so on in some scrambled order only the setter knows. What's left looks like a spreadsheet that has given up. It is also, remarkably, still solvable, and that is the entire trick of the codeword.

• 26 letters in every grid — all of them, guaranteed, including the awkward ones
• 9,272 hand-vetted words the grids are built from, three to ten letters
• 61 grid shapes, each filled fresh in your browser — no two evenings alike

How it plays

Pick a cell and type a letter. Because a codeword's unit is the number, not the square, that letter lands in every square sharing that number at once — which is what makes the puzzle cascade. The key panel beside the grid tracks all 26 numbers as you assign them, and the alphabet strip underneath crosses letters off as you use them, so you can always see what's left to place. On a phone, tapping a letter in that strip types it; on a keyboard, just type.

Check marks anything wrong in red and costs you nothing. Reveal is there when a grid beats you, but it voids that solve — the stats on this site only ever count honest ones. A timer runs quietly, each difficulty keeps its own best, and a finished grid can be shared as a challenge link that deals a friend the identical code.

Where it came from

The codeword is not really a descendant of the crossword; it is a descendant of the cipher. Substituting one letter for another is the oldest trick in cryptography — Suetonius describes Julius Caesar shifting each letter three places along the alphabet to keep his correspondence private — and for centuries breaking such a substitution was serious work rather than a pastime. What the codeword adds is scaffolding. A plain substitution cipher gives you a passage of text and wishes you luck; a codeword gives you the same puzzle laid out in a grid where every word crosses two or three others, so a single correct guess is immediately tested from several directions at once.

It surfaced as a newspaper and magazine fixture in Britain, where it is printed under a small pile of names — Codeword, Code Cracker, Kaidoku, Cipher Crossword — depending on who is publishing it. The conventions settled early and have barely moved since: a 13×13 grid, numbers 1 to 26, every letter of the alphabet somewhere in the grid, and two or three letters given away at the start. This page follows all of them, because after fifty years of testing they turn out to be right.

What the letters in these grids actually do

Solvers are usually handed the textbook frequency order — ETAOIN SHRDLU, the letters of English ranked by how often they turn up in running prose. That ordering comes from continuous text, though, and a codeword contains no continuous text: it contains 9,272 candidate words, with no the, no of, and no and repeating endlessly to weight the count. So here is the real thing, measured across the exact word list these grids are built from.

• Your working shortlist
• E 11.5%
• A 8.6%
• S 8.2%
• R 7.5%
• O 6.8%
• T 6.3%
• I 6.2%
• N 5.8%
• L 5.8%
• C 3.7%
• The long tail — where the loners hide
• D 3.7%
• P 3.3%
• U 3.2%
• H 2.9%
• M 2.9%
• G 2.7%
• B 2.4%
• W 1.7%
• Y 1.7%
• F 1.6%
• K 1.6%
• V 1.0%
• Z 0.4%
• J 0.3%
• X 0.3%
• Q 0.2%

Note where S has landed. In prose it sits mid-table; in a word list it is inflated by every plural and every third-person verb, and here it is the third commonest letter of all. At the other end, Q · X · J · Z · V between them barely register — Q at 0.2% against E's 11.5%.

Before writing anything, count how often each number occurs. The busiest three or four are your vowels and your S; the loners are where the alphabet's awkward squad is hiding.

Five tells that break a grid open

Every one of these is a shortcut past guessing — and every number in them is counted from the same word list the grids are filled from.

• 6__ Start where the starters are Fill the given letters into every square carrying their numbers first, then find the word with the most squares already filled. That word is the puzzle's weakest point, and where the first cascade begins.
• QU Q drags a U behind it The one piece of received wisdom that survives contact with the data: of the 85 words here containing a Q, exactly 1 manages without a U — QATAR, a place name. Find the letter that only ever appears immediately before the same partner and you have found them both.
• OO Doubles are a short list Two identical numbers side by side narrow English sharply: the commonest doubled letters here are OO, LL, EE, SS, in that order. A doubled letter is rarely a vowel other than O or E, and essentially never one of the rare five.
• __S Watch the last square 20% of the words here end in S. If one number keeps turning up in the final square of word after word, you are almost certainly looking at S — and confirming it typically unlocks a dozen squares in one stroke.
• 3 Mine the three-letter words A three-square answer feels like too little to work with. It's the opposite: short words are the most constrained things in the grid. Once two of the three squares are known, usually exactly one English word fits. Sweep them every time you place a letter.

How these grids are built

Two things have to be true before a codeword is worth anyone's evening, and neither happens by accident. Every grid you are dealt has been through all three of these steps before it reaches the page.

1. Hunt the missing letters. Left to itself, a filled grid this size averages around twenty-one distinct letters — J, Q, X and Z simply never show up, because the words carrying them are rare and the fill has no reason to reach for one. So the fill is deliberately greedy: among the words that would fit a slot, it tries the ones bringing letters the grid still lacks first, weighted by how scarce that letter is. It only reorders candidates, never removes any, so it cannot make a fillable grid unfillable — and it takes the pangram hit rate from zero to most attempts.
2. Prove it can be deduced. Every grid is handed to a solver of our own first. That solver knows only what you know — the numbers, the starters, and that every entry is a real word — and it narrows each slot until some number's letter is forced by every remaining candidate agreeing on it. Because each step is forced rather than guessed, finishing proves something stronger: the grid has exactly one solution. If the solver can't crack it, you never see it.
3. Set the lighting. The starter letters fall out of the same process — the fewest the solver needed, topped up to what your difficulty advertises, favouring the numbers that fill the most squares. An easy grid isn't a smaller puzzle so much as a better-lit one, and a hard grid is the same machinery with the lights turned down.

At no point are you asked to guess and hope. If a grid is on your screen, a solver has already proved it can be reasoned out.

On paper

Codewords were a print puzzle long before they were a screen one, and they lose nothing in the move back. The pack button above builds a PDF of fresh grids — never ones you have already played here — with the code key laid out for pencil work, an optional name-and-date line for classrooms, a large-print option, and an answer key that includes the full number-to-letter map rather than just the finished grid, since the code is the part you actually spent the evening on. Teachers get a bonus the puzzle didn't set out to provide: a codeword is a spelling and letter-pattern exercise wearing a disguise, and children who will not sit still for a worksheet will happily fight a grid.

If you're enjoying this

The obvious next stop is a proper clued grid — the endless mini crosswords or the daily daily crossword, both built by the same engine that fills the grids above. If it was the letter-frequency detective work you enjoyed rather than the crossword shape, the word guessing game runs on exactly that instinct, and anagrams ask you to rearrange letters rather than decode them. Everything here is free, needs no account, and has no daily ration.