Crossword Solver

Crossword puzzles give you two pieces of information: the clue (a definition, wordplay hint, or trivia question) and the grid (a fixed number of squares with some letters already filled in). A crossword solver combines those constraints to propose answers from a dictionary. Our implementation uses two distinct search modes because real solvers think in patterns or clues — rarely both at once in the same mental step.

In pattern mode, you type the answer mask exactly as it appears in the grid. Known letters stay as letters; blank squares become question marks or underscores. The engine parses your mask into fixed positions, selects the matching length bucket from our pre-built dictionary index, and filters candidates in milliseconds — no full-dictionary scan when length is known.

In clue mode, you paste or type the clue verbatim. We tokenize the clue, score similarity against a clue corpus (definitions, synonyms, and tags), and rank answers with a confidence percentage. Optional length filters disambiguate clues like "Japanese capital (3)" where multiple cities might partially match.

Every result shows word length, Scrabble score, and a definition preview when available. Click any answer to open its word detail page with anagrams, related words, and links back to other tools on unscrambletools.com.

Crossword patterns are positional constraints. Unlike a "contains QU" filter, pattern ?R?W? requires R in the second position and W in the fourth — nothing else qualifies. This is the same logic Wordle uses for green tiles, but crossword patterns are simpler: there is no yellow "wrong position" feedback, only known letters and blanks.

We accept both ? and _ as wildcards because different publications and solver apps use different conventions. Mixed input like H__S_N normalizes internally to a single wildcard character before matching.

Optional filters stack on top of the pattern: starts with and ends with for partial theme answers, contains when you know a substring must appear somewhere outside fixed positions, and exclude letters when crossing entries rule out certain characters. These filters mirror how experienced solvers eliminate impossible candidates after checking perpendicular words.

Example: pattern A?P?E at five letters matches APPLE, AMPLE, and other words sharing A in column one, P in column three, and E in column five. Adding exclude letters z removes obscure variants you already know are wrong from crossing letters.

Clue solving is harder than pattern matching because clues are written for humans, not databases. A clue like "Lead actor?" might mean STAR, HEAD, or even PB (chemical symbol) depending on wordplay. Our clue engine uses layered matching: exact and fuzzy phrase similarity, token overlap with stop-word removal, synonym fields in the clue corpus, and lightweight metadata fallback when no curated clue exists.

Each clue result includes a confidence score from 0–100%. High confidence (80%+) usually means a strong match in the seed clue database. Mid-range scores suggest partial token overlap — worth checking against crossing letters. Low scores may be metadata-only matches; treat them as starting points, not definitive answers.

The clue architecture loads from data/dictionaries/crossword/clues.seed.json when present, falling back to an embedded seed list. This lets us swap in a full crossword clue database later without changing the API or UI — only the data file and import pipeline.

Pattern A?P?E (5 letters): Matches common five-letter words with A first, P third, E last — typical when crossings give you A at 1-Across and E at 5-Across. Top results include APPLE and AMPLE; check crossing down words to pick the right one.

Pattern ?R?W? (5 letters): Finds words with R in position two and W in position four — useful for partial fills like CROWD or CROWN once C or N is confirmed by another entry.

Clue "greek god of war": Returns ARES with high confidence from the mythology clue corpus. Cross-check length: four letters fits most daily puzzles; MARS is the Roman equivalent if the grid demands four letters starting with M instead.

Clue "capital of italy": Returns ROME with strong token overlap. Geography clues map well to the seed database; cryptic clues with anagram indicators may need pattern mode instead.

Pattern Finder excels at exploratory searches — all five-letter words starting with RE, words containing QU, six-letter words with X but not Z. It does not treat per-position masks as first-class input. Use Pattern Finder for vocabulary building; use Crossword Solver when you have a grid mask.

Anagram Solver rearranges a fixed set of letters from a rack or scrambled phrase. Crossword solving rarely gives you a bag of letters — it gives you fixed positions and blanks. Anagram mode helps cryptic crosswords with anagram indicators ("mixed up", "broken") but not standard grid fill.

Wordle Solver handles green/yellow/gray feedback across multiple guess rows for five-letter words only. Crossword patterns generalize to any length and do not model yellow "wrong position" constraints — unless you encode them manually as known letters in other positions.

Word Unscrambler remains the best choice for Scrabble and Words with Friends racks with wildcard tiles. Crossword Solver links to it (and every word page) when you need related tools after finding a candidate answer.