Sources and licences

Every source this edition was built from, with its licence. Last updated September 20, 2026

Journey to the West is operated by Technologies InstantHPI Inc. The list below is the record kept for this title; a licence says what may be done with the material it covers, and the material is used here under that licence.

Sources — Journey to the West, short (西遊記精選)

Everything here is free to use. Nothing in this title comes from a paid image or video API. The passage text is public domain, the dictionary and the stroke data carry open licences, and the passages' English and French renderings as well as the word glosses are machine-assisted (DeepSeek text API) and reviewed by a human before release, as the brief asks.

Passage text — 西遊記, Wu Cheng'en (Ming dynasty), public domain

• Source: Chinese Wikisource, https://zh.wikisource.org/wiki/西遊記 (index page), chapter pages

西遊記/第001回 … 西遊記/第100回, read through the MediaWiki API (https://zh.wikisource.org/w/api.php).

• Licence: the novel is public domain (author died 1582; the Wikisource transcription is released

under CC BY-SA 4.0 / public domain as the page states). Wikisource asks for attribution and share-alike for the transcription; we credit Wikisource in this file and in the app's about panel.

• Taken: the raw wikitext of all 100 chapters plus the index page (for the chapter couplets), the

cleaned chapter text, and the 365 passages cut verbatim out of it. Each passage keeps srcOffset, its character offset in the cleaned chapter text, so the cut can be proved: cleanChapter(raw).plain.slice(srcOffset, srcOffset + sa.length) === sa for all 365 passages (the offset counts JavaScript string units, as slice does; the builder proves it on every run and the proof is printed by the verify stage).

• Fetched: 2026-10-01.
• Cache: _scripts/.cache/wikisource/index-wikitext.txt, _scripts/.cache/wikisource/api-NNN.json,

_scripts/.cache/wikisource/chapter-NNN.txt (verbatim server answers).

• Cleaning: removed the {{header}}/{{footer}} templates, resolved {{另|a|b}} to a,

{{參|c|note}} to c, dropped editorial templates (Pd-old, PD-old, CJK-New-Char) and the -{zh-hans:…;zh-hant:…}- converter markup (the traditional branch is kept), then stripped links and HTML entities. No word of the novel is changed.

Dictionary — CC-CEDICT

• Source: https://www.mdbg.net/chinese/export/cedict/cedict_1_0_ts_utf-8_mdbg.txt.gz
• Licence: CC BY-SA 4.0 (stated on the download page and in the file header); attribution to

CC-CEDICT and its maintainers is given here and in the app's about panel.

• Taken: the headword pairs (traditional, simplified), the pinyin and the English senses. Word

segmentation of every passage is verified against this dictionary: every multi-character word is one of its headwords (or a proper name from _scripts/names.json), and a word it does not know is split to single characters. The gloss of a word comes from the dictionary entry whose pinyin matches the pinyin-pro reading of the word, and a sense that is only a surname or only a cross-reference ("variant of …", "see …", "same as …") is never used; where nothing usable is left the cross-reference itself is followed for a meaning.

• Fetched: 2026-10-01. Cache: _scripts/.cache/cedict/cedict.txt.gz.

Character frequency — counted by this project from the texts the series teaches

• Source: the complete 西遊記 (Wu Cheng'en, Ming dynasty) from Chinese Wikisource, the Wang Bi 道德經

and Xuanzang's 心經, the same public-domain texts the titles teach.

• Counted by: work/han-frequency.cjs, which walks those texts, keeps every Han character (Unicode

script Han; no punctuation, Latin or digits) and ranks them by how often they occur, the most often first. No third party's list is used and there is no licence to obey; the texts are public domain and the count is ours.

• Taken: the rank of every character in the 365 passages, mapped through the CC-CEDICT

traditional/simplified pairs; rank on every word is the rank of the rarest character in the word, because a word can only be read once its hardest character is known.

• Built: node work/han-frequency.cjs. File: work/frequency/han-pd.json.

Stroke data — Make Me a Hanzi

• Source: https://github.com/skishore/makemeahanzi (graphics.txt)
• Licence: LGPL-3.0 for the code and Arphic Public License for the font outlines derived from

Arphic's fonts; both are named in title.json and here.

• Taken: the outline and median of every stroke of every character in the 365 passages, written to

data/strokes/<hex>.json in Hanzi Writer format, one file per character.

• Fetched: 2026-10-01. Cache: _scripts/.cache/hanzi/graphics.txt (30,778,076 bytes).

Font — Noto Serif TC

• Source: https://github.com/google/fonts/tree/main/ofl/notoseriftc (variable font

NotoSerifTC[wght].ttf, fetched 2026-10-01; cache _scripts/.cache/font/NotoSerifTC[wght].ttf).

• Licence: SIL Open Font License 1.1; the licence text ships beside the font as

assets/fonts/OFL.txt.

• Taken: the Regular instance, subset to the characters this title writes and saved as

assets/fonts/NotoSerifTC.woff2 (_scripts/subset-font.py, fontTools).

Segmenter — jieba-wasm

• Source: https://www.npmjs.com/package/jieba-wasm. Licence: MIT.
• Taken: default-mode Chinese segmentation of every passage.
• Installed: 2026-10-01. Cache: _scripts/node_modules/.

Pinyin — pinyin-pro

• Source: https://www.npmjs.com/package/pinyin-pro. Licence: MIT.
• Taken: pinyin with tone marks for every word and every passage (tl).
• Installed: 2026-10-01. Cache: _scripts/node_modules/.

Translations (English and French) — DeepSeek

• Source: DeepSeek text API, model deepseek-flash, https://api.deepseek.com/chat/completions
• Licence: the translations are produced for this project and are released with the title; the API

terms allow the use of the output. Named "InstantHPI (2026), machine-assisted" in data/languages.json, exactly as the brief asks. No public-domain English translation follows the novel passage by passage (Yu and Jenner are in copyright; Timothy Richard 1913 is an abridged paraphrase), so every tr in English and French, the English gloss of every word, the French gloss of every word and the English rendering of each chapter couplet is machine-assisted; the text is reviewed by a human before release.

• The English gloss of a word is the meaning it carries in its passage: ten passages to a call, at

temperature 0, given the passage translation and the dictionary senses of each word, answered as one numbered line per word. A batch that answers badly is asked once more and otherwise keeps the dictionary sense; nothing is paid twice, because every request and answer is cached by the sha1 of the request under _scripts/.cache/glosses/, and the run's counts are in _scripts/.cache/gloss-report.json.

• Fetched: 2026-10-01. Cache: _scripts/.cache/deepseek/ (one JSON answer per call, verbatim).

Contact

Questions about a source or a licence? Email info@instanthpi.ai.