How it works

The search

The 172,749-word lexicon is stored as one file per word length. When you search, only the lengths that can possibly match are fetched: a seven-letter rack needs lengths two through seven, which is roughly 51,000 words, not the whole list. Files are cached after the first fetch, so a second search of the same size costs nothing.

Matching itself is a letter-count comparison. Each word is reduced to a 26-slot vector of how many times each letter appears, and a word is spellable if every count fits inside your rack — with any shortfall covered by a blank, up to the number of blanks you have. That is exact, not heuristic: there are no near-misses and no ranking guesswork.

Crossword patterns work differently. There the position of each character matters, so the pattern fixes the word length and each square is compared individually.

Scoring

A word's score is the sum of its tile values under whichever game you picked. Two things are deliberately excluded:

• Board multipliers. Double-letter and triple-word squares depend on where the word lands, which this site cannot see. A bonus-inclusive number would be a guess presented as a fact, so the figure shown is the face value you can verify by adding up the tiles in your hand.
• Optimistic blank scoring. A blank is worth zero, so a word that needs one has the value of its most expensive letter removed before totalling. Spelling quartz with a real Q is 24; with a blank Q it is 14. Tools that skip this step overstate what you will actually score.

The bingo bonus for using all seven tiles is added where it applies — 50 in Scrabble, 40 in Wordfeud, 35 in Words With Friends — and those words are highlighted.

The Wordle filter

Green fixes a letter to a position, yellow says the letter is present but not there, and grey caps how many times it may appear. The subtlety is duplicates: the game colours a guess in two passes, greens first and then yellows against the answer's remaining letters, so a repeated letter you guessed more often than the answer contains comes back grey on the extras.

That means grey does not mean "absent" — it means "no more than the number already marked". Our filter derives a minimum and a maximum count per letter from each row rather than treating grey as elimination, which is why it does not discard the right answer when you have guessed a doubled letter.

Word search generation

Words are placed longest-first into a grid, each one tried at random positions and directions until it fits without contradicting a letter already placed. Overlaps are allowed and encouraged — they make a denser grid. Remaining cells are filled with random letters.

Placement can fail, and when it does the page says which words could not be fitted and why, rather than silently dropping them. Every puzzle carries a seed, so the same seed and settings reproduce the same grid exactly.

Crossword generation

The mini crossword grids are filled by backtracking: rows are placed one at a time, and each column carries a live position in a prefix tree of the word pool, so a candidate is abandoned the moment any column stops being the start of a real word.

Two constraints shape the result. The word pool is restricted to common words that also have a definition on record, so every entry is both recognisable and clueable. And the black squares are placed off the main diagonal — with a symmetric pattern the filler produces word squares, where the across and down answers are the same five words and half the puzzle solves itself.

Syllable counting

Vowel groups, plus the standard exceptions: silent final E, -LE endings after a consonant, -ED and -ES that do not add a syllable, and the diphthongs that look like two vowels and are pronounced as one.

This is the one genuinely approximate tool on the site. It handles ordinary vocabulary well and will disagree with you on loanwords, names, and the words where speakers themselves disagree — fire, hour and every are one or two syllables depending on the accent, and no algorithm settles that. Trust your ear over the number.

Word-list pages

The word lists are generated at build time from the same lexicon. Counts, groupings, most-common words and highest-scoring words are all computed per page from the actual data rather than templated.

Combinations with fewer than fifteen words do not get a page. "Nine letter words starting with X" has a handful of entries, and a page with a handful of entries is a bad answer to a search — so it is left out rather than padded.

What none of it does

No search on this site reaches a server. There is no query log, no personalisation and no model deciding what you meant — just a dictionary in your browser and an exact comparison against it. That is a limit as well as a feature: nothing here corrects a typo for you or guesses at what you were trying to type.