WORDWARD FRENCH VOCABULARY — DECK VERSION 1
2026-10-24

Source: Grammalecte / Dicollecte, Olivier R., Lexique des formes fléchies du français v7.7 (2025).
Official project: https://www.grammalecte.net/
Exact archive: https://www.grammalecte.net/dic/lexique-grammalecte-fr-v7.7.zip
Archive SHA-256: bdccf2065e252010253171000a38de84581407896b6bc60aca7d4d69dab7e4d2
License: Mozilla Public License 2.0, as stated in the archive's README_lexique.txt.
Full license included in MPL-2.0.txt, fetched from https://www.mozilla.org/media/MPL/2.0/index.txt
The original archive, original notice and explicit answer-selection source are included.
Derived word data are MPL-2.0; this does not change unrelated game-code licensing.

REPRODUCIBLE BUILD
python scripts/build-french-words.py
The standard-library generator reads the pinned archive offline, asserts its hash,
20-column schema and 515,693 record count, and checks every answer's exact spelling.
It generates assets/words-fr.txt, src/french-words.js, a readable license notice,
and validation-report.json with per-length counts and content hashes.

PLAYABLE FORM
NFC uppercase. Explicit French accents and cedilla fold to A-Z. OE and AE ligatures
expand into two cells. Only 3..7 expanded cells qualify. The selected original
spelling is retained for solved/revealed answers. Normalized duplicates share one
answer slot. No arbitrary Unicode-mark stripping and no N/Ñ equivalence.
This matching convention is explained in help; it is not a spelling recommendation.
Orthography references:
https://www.academie-francaise.fr/questions-de-langue
https://vitrinelinguistique.oqlf.gouv.qc.ca/21611/lorthographe/problemes-lies-aux-voyelles/ecriture-et-prononciation-des-lettres-doubles-ae-et-oe

GUESS DICTIONARY
57,217 normalized guesses. Original lower-case word forms, including plurals,
conjugations, loans and traditional/reformed spellings. Capitalized entries,
proper-name/acronym tags, punctuation and multiword expressions are excluded.
Homographs of names remain ordinary words where the source also supplies them.
This broad spelling lexicon is not used wholesale as an answer list. Source
inclusion proves dictionary presence, not everyday familiarity or completeness.
Examples verified in the source include CHUM, DRACHE and NONANTE; this is not a
claim of complete regional coverage. Eight-cell words such as SEPTANTE are outside
the current seven-cell maximum even though valid French words.

ANSWER SELECTION
3,043 explicit spellings in answers-fr.txt. A candidate pass selected 2,801
normalized words with source frequency index >= 7, no source Notes, and lemma or
non-verb-inflection status. Each resulting candidate list was inspected. Selection
removed corpus fragments, abbreviations and some specialist/archaic prompts, then
added verified everyday words. selection-review.json lists removals and additions.
Its unverifiedNotAdded entries refer to display spellings absent from the filtered
source, not rejection of their normalized guesses: BOEUF, CANOE and NOEL are accepted.
The final explicit answer file, not frequency thresholds, is the versioned source
of truth. Order determines the daily stream and must be versioned if it changes.
No word is weighted or chosen according to the player's health, skill or luck.

ASSISTANT REVIEW, NOT NATIVE SIGN-OFF
Vocabulary and interface text have assistant review only. Familiarity across
France, Canada, Belgium, Switzerland and other Francophone communities needs
native players. Three-letter forms such as BIS, DOT, DUS and UNS and literary
vocabulary deserve focused review; a valid dictionary entry can still be a poor
answer for some audiences. No existing game's answer deck or implementation was
copied. French Wordle precedent inspected: https://wordle.louan.me/ and the author's
https://github.com/LouanBen/wordle-fr .

SIMULATION
A separate wordfreq 3.1.1 prior is built only from the accepted dictionary. It does
not ingest the selected answer file. Its data are CC BY-SA 4.0, with separate
attribution in scripts/simulation/data/NOTICE.txt. Corpus frequency is a hypothesis
about familiarity; it is not measured French-player performance.
