# Word list licenses (commercial use)

This site uses **only** word-list sources that are free to use commercially
(public domain, Unlicense, MIT, MPL-2.0, or permissively licensed
redistributions of public lists). Every count below was measured from the files
actually shipped in this build (**2026-10-18**), not from build-time estimates.

## Sources merged

| Source | License | Use |
|--------|---------|-----|
| dwyl/english-words (`words_alpha.txt`) | Unlicense | Commercial OK |
| dwyl/english-english-words (`words.txt`) | Unlicense | Commercial OK |
| dwyl/english-words (`word_list_moby_*`) — Moby Words II, Grady Ward | Public Domain (via Unlicense repo) | Commercial OK |
| ENABLE-style base (public-domain tradition) | Public Domain | Commercial OK |
| Aggregated public lists via MIT wordlist repos (SIL, corncob, umich, etc.) | Public lists / MIT aggregator | Commercial OK |

### Provenance caveat

The `dwyl/english-words` README states that `words.txt` was extracted from an
infochimps word list and that "Copyright still belongs to them". The repository
itself is released under the Unlicense, which is why we treat it as freely
redistributable, but the caveat is recorded here rather than hidden.

## Not used

- Official Scrabble tournament lists (TWL/NWL/CSW) — proprietary
- Merriam-Webster / Oxford full text — proprietary
- Unclear or non-commercial-only dumps
- **GNU Aspell dictionaries (aspell-de, aspell-es, aspell-fr, aspell-it).** These
  are excellent lexicons and were evaluated for this build, but every one of
  them is licensed **GPL-2-or-later / GPL**, which cannot be relicensed into a
  permissively licensed or ShareAlike word list without creating a licensing
  conflict. They were deliberately left out. The MPL-2.0 option offered by the
  Hunspell dictionaries below was used instead.

## Files shipped in this build

| File | Role | Words |
|------|------|-------|
| `words.txt` / `words.js` | **Standard** tool default list | **168,479** |
| `words-extended.js` | **Extended** optional wider list | **1,341,495** |

Tools default to **Standard**. Extended is optional and noisier.

Extended is a **strict superset** of Standard: all 168,479 Standard entries are
present in Extended, plus 1,173,016 additional forms. Counting English coverage
once therefore yields 1,341,495 unique English forms, not a sum of the two files.
(An earlier build was 17 entries short of a true superset; those 17 Standard
forms are now present in Extended.)

## The `[a-z]` language lock

Both English files are now filtered to **lowercase ASCII letters only, 2-24
characters** (`^[a-z]{2,24}$`).

This build removed **84 tokens from Standard** and **55,143 from Extended**
that violated that rule. They were of three kinds:

1. **Quote fragments.** The Extended file contained strings such as `"aa`,
   `"and`, `",` — leftover pieces of subtitle or CSV text, not words.
2. **Non-English spellings.** `aalbæk` (Danish), `aandrã` (Afrikaans),
   `aartín` (Galician/Spanish), `aavikkopläntin` (Finnish), `abachitdzé`
   (Chuvash) and similar forms had entered the Extended merge.
3. **Diacritics.** Loanwords with accents (`café`, `naïve`) were dropped too,
   because no English tool on this site can match them: every pattern filter,
   rack matcher and character counter in `assets/` is ASCII-only.

No English tool could ever have returned any of those tokens, so nothing that
worked before stops working. What is removed is dead weight that also made the
list measurably *not* an English list.

## Related coverage figures

| Layer | Count |
|-------|-------|
| Standard list (`words.js`, `words.txt`) | 168,479 |
| Extended list (`words-extended.js`) | 1,341,495 |
| Dictionary headwords (`assets/dict/`) | 345,078 |
| Union of English spellings across all three layers | **1,341,495** |
| Definition senses (`assets/dict/`) | 743,830 |

Of the 345,078 dictionary headwords, 345,047 already appear in Extended and only
31 are dictionary-only. Definition senses total 743,830: **542,299** from
English Wiktionary (CC BY-SA 4.0) and **201,526** from Princeton WordNet
(WordNet 3.0 licence), plus 5 senses with no source label. See `credits` and
`LANGUAGE-LICENSES.txt`.

`assets/starts/*.json` (the A–Z browse shards) is derived from the Standard
list and was regenerated in this build: 168,479 entries across 26 shards.

### Dictionary headword cleanup

An earlier build stripped spaces from Wiktionary and WordNet multi-word entries
while sharding, which turned phrases into nonsense headwords:
`abnormal psychology` became `abnormalpsychology`, `above the salt` became
`abovethesalt`, `Genista schottiana` became `genistaschottiana`. **20,057** such
mangled entries were removed from `assets/dict/` in this build (headwords went
from 365,135 to 345,078, senses from 764,347 to 743,830). Only entries that are
not in Standard or Extended *and* can be re-split into two or more English
words were removed, so no genuine single-word headword was lost.

### Dictionary data cleanup

100,894 senses stored a broken etymology value — the JavaScript string
`"[object Object]"` produced by an upstream importer bug. The dictionary UI
renders that field as the "Example" line, so visitors were seeing literal
`[object Object]` text. All 100,894 were cleared in this build.

## Not shipped

- `words-commercial-full.txt` — not included in this package. License notes
  above describe open sources used to build the lists that **are** shipped.

Tools always default to Standard for trustworthy results. Extended may include
rare or noisy forms.