Crossword solver · ENABLE1, 172,820 words · length locked by the pattern
Every word that fits the squares you have
Type the letters you are sure of and an underscore for each square you are not. The number of squares is the number of letters, so the sieve reads one slice of a named public-domain dictionary — every word of that length — and shows the ones that fit. It counts what it read, and says so when nothing fits.
One character per square: a letter you know, or _ (or .) for one you do not; spaces are ignored. ? is the blank tile on the rack pages, not a square here.
The word index is not loaded yet. It ships as 14 plain-text shards, one per word length, totalling 3,204,287 bytes; a pattern locks its own length, so exactly one of them is pulled, and the figure for yours appears here after a fit.
What is measured, how it is kept, and what it cannot tell you
What is measured. One thing: which entries of a named word list are exactly as long as your pattern and carry each letter you gave in exactly the square you gave it. A blank square is not a wildcard for "any number of letters" — it is one square, one letter. There is no ranking and no guess at which answer is wanted: a word fits or it does not, and the count in the strip is the size of that set, not a sample of it.
How it is kept. The list ships as a file, wordlist/enable1.txt, precomputed at build time into an index split by word length. A positional question cannot be answered from the anagram keys that index is built on — sorting a word's letters throws its positions away — so this page reads the one shard whose length your pattern locks and tests each word in it. That is a scan, named as a scan on the methodology page, bounded by the words of a single length instead of all 172,820: the strip prints how many were read, and the line under the results the bytes pulled.
What it cannot tell you. What the clue means, or which answer the setter wanted — nothing here reads clues. Whether the crossing letters you typed are the right ones; a wrong one produces a confident, empty answer. And anything about words outside ENABLE1: proper nouns, phrases and abbreviations are absent from the list, and crossword answers are often exactly those — the honest limit of matching a general word list against a grid.
The pattern locks the length, so one slice is the whole search
Most of the work a solver does is deciding what it can ignore, and here the pattern does it: nine squares can only be filled by a nine-letter word, so the eight- and ten-letter words are not searched at all. The index ships as one plain-text file per word length and a fit pulls exactly one of them, which is why the line under the results names a single shard and why re-fitting at the same length fetches nothing. A pattern of nothing but blanks is a legal question — every word of that length — and is answered rather than refused, capped for painting with the true total printed beside the cap. One longer than the longest indexed word is refused by name instead, because no word in the dictionary could fill it.