WORDWARD BRAZILIAN PORTUGUESE VOCABULARY — DECK VERSION 1
2026-10-25

Integration note: the browser adapter uppercases only ASCII and explicitly
supported Portuguese letters, then folds their accents/cedilla. Unsupported
dotless I, long S, sharp S and presentation ligatures remain invalid rather than
expanding into another guess. Pasted punctuation stays visible for rejection.
This strengthens input handling without changing the generated vocabulary.

SOURCE
VERO - Verificador Ortográfico Livre, Raimundo Santos Moura and the Brazilian
LibreOffice community. Copyright (C) 2006-2013 Raimundo Santos Moura.
The pinned affix file identifies itself as VERO 3.2. This is the pt_BR source
distributed by LibreOffice; it is not a claim that every contemporary word is in it.
https://github.com/LibreOffice/dictionaries/tree/31bc2a1104a1cd175f900902f994c76dea35c763/pt_BR
Exact files: pt_BR.aff, pt_BR.dic, README_en.txt and README_pt_BR.txt.
The generator pins each source SHA-256; validation-report.json repeats them.
Original files are included without alteration, including their UTF-8 byte marks.

LICENSE
The original notices expressly offer LGPL version 3 and Mozilla Public License.
Because the old MPL notice does not identify a version, this adaptation chooses
the unambiguous LGPL version 3 option. The derived word-data files, including the
selected answer deck, are LGPL-3.0-only. COPYING.LESSER and its incorporated
GPLv3 terms in COPYING are included, downloaded from:
https://www.gnu.org/licenses/lgpl-3.0.txt
https://www.gnu.org/licenses/gpl-3.0.txt
The original author notices and modifications are repeated in the public
assets/word-list-pt-license.txt. No definitions or another game's answer deck
were copied. These data terms do not change unrelated game-code licensing.

REPRODUCIBLE BUILD
python scripts/build-portuguese-words.py
Python standard library only; offline; no frequency dependency during builds.
The generator checks all pinned hashes, the 312,368-row dictionary count, the
103 affix classes and their rule counts, and the supported Hunspell directives.
It expands one suffix and crossed prefixes. This pinned source has no second
morphological suffix continuations: its only continuation flag is NOSUGGEST on
state-name expansion rules. Unknown features fail rather than silently dropping
words. Forbidden spellings are removed even if generated from a different root.
Only originally lower-case Portuguese-letter roots enter the guess lexicon;
proper-name capitalization, punctuation and multiword expressions are excluded.
Accepted plurals and conjugations are expanded. No guesses are constructed by
inventing a Portuguese ending without an attested source affix rule.

PLAYABLE FORM
NFC; preserve unsupported sharp S as capital sharp S before uppercasing; explicit
ÁÀÂÃÉÊÍÓÔÕÚÜÇ -> AAAAEEIOOOUUC; retain 3..7 A-Z cells. Unsupported letters are not
silently converted to unrelated letters. This is a game matching convention,
not a spelling recommendation. Correct chosen answer spelling is shown on solve.
Foreign names/words requiring characters outside this set are outside this deck.
Portuguese Wordle precedent inspected: https://term.ooo/ explains automatic
accents and accent-insensitive clues. No code or answer data from it are reused.

COLLISIONS
169,108 unique normalized guesses come from 185,841 original spellings. The
16,733 reduction counts extra spellings collapsed, not distinct collision groups.
AVÓ/AVÔ, TEM/TÊM, VEM/VÊM and POR/PÔR share a playable identity under this policy.
Each answer identity occurs once. Selection-review.json records encountered
answer collisions and explicit display choices; the same spelling is always
shown for a given selected answer. Both accepted input spellings solve that board.
Unaccented input and an old accented variant may normalize to a currently valid
board word; this game does not claim to reject every orthographic error.

ANSWER SELECTION
2,902 explicit spellings in answers-pt.txt: 148 three-letter, 426 four-letter,
865 five-letter, 795 six-letter and 668 seven-letter entries.
2,567 initial source-attested candidates had wordfreq 3.1.1 Portuguese Zipf >=4.3.
The assistant read those lists, removed listed proper names, abbreviations,
web/corpus artifacts and some unsuitable prompts; then added everyday animals,
food, household items, natural features and basic verbs verified in VERO.
Inflections and common function words remain answers; the answer policy is not
nouns-only. The final explicit file is authoritative. Reordering/changing it needs
a deck-version decision because it affects deterministic daily streams.

The optional review-selection.py recreates the recorded editorial process using
wordfreq==3.1.1 in .qa/simulation-python. The upstream Portuguese frequency table
is used only as a candidate aid, and contains mixed regional corpora. VERO pt_BR
is the spelling authority for this deck; this is not a separate European
Portuguese edition. Regional vocabulary completeness is not claimed.
wordfreq: Robyn Speer, https://pypi.org/project/wordfreq/3.1.1/; frequency data
CC-BY-SA-4.0. No frequency scores are included in runtime word assets. The game's
independent simulation prior has its own attribution and is not this answer list.

VERIFICATION AND LIMITS
The optional verify-expansion.py uses spylls==0.1.7 in .qa/german/python and
wordfreq==3.1.1 in .qa/simulation-python. Independent spelling validation checked
every one of the 185,841 expanded original spellings: zero rejected. It checked
87,252 corpus candidates outside the expansion: zero additional accepted forms.
Those counts establish source expansion soundness and a bounded completeness
check, not that every Portuguese word is accepted or that every answer is fun.
Positive and negative probes establish that the validator is operating; full
reports include their names and scope. The normal build also detects seven
deliberately injected faults after an unmodified positive baseline, including
unattested/duplicate/unsupported answers, an empty deck, an unknown affix feature,
and a missing forbidden-word policy.

Native-speaker review remains pending. This is assistant editorial review only.
Brazilian players should especially review three-letter function/verb forms,
accent-colliding meanings, regional terms and the proportion of conjugations.
Corpus frequency and successful simulation do not replace that review.
