Everything we derived from open data is here, one file per source, because their licenses cannot be merged into one file. Each file opens with a header naming its license, source, version and what we changed (the Arphic Public License and the share-alike licenses ask for exactly that). Data date: 2026-10-13.
Files
| File | License | What is in it | Size |
|---|---|---|---|
| ja-strokes.json.gz | CC BY-SA 3.0 (KanjiVG); adaptation | KanjiVG stroke paths for kanji and kana as used on StrokeFox: one record per character, paths without style attributes, stroke-number positions, component groups. | 3.1 MB |
| zh-strokes.json.gz | Arphic Public License (Make Me a Hanzi graphics), modified | Make Me a Hanzi outlines and medians as used on StrokeFox: whitespace normalised, medians rounded and flipped to y-down (y' = 900 - y), Q curves converted to C in the PDF writer. | 12.4 MB |
| zh-char-info.json.gz | LGPL-3.0 (Make Me a Hanzi dictionary), modified | Pinyin, definition, decomposition, radical, etymology and stroke-to-part matches from the Make Me a Hanzi dictionary, reduced to the fields StrokeFox shows. | 333 KB |
| ja-words.json.gz | CC BY-SA 4.0 (EDRDG licence) | Per kanji: readings and meanings from KANJIDIC2 and up to 12 common JMdict words with reading and gloss; per kana: common words containing it. | 711 KB |
| zh-words.json.gz | CC BY-SA 4.0 (CC-CEDICT) | Per hanzi: its readings, most used first, each with cleaned glosses and traditional forms (Unihan kTraditionalVariant), up to 12 CC-CEDICT words; senses unfit for learners removed (tools/safety.py); measure-word counts. | 1.6 MB |
| levels.json.gz | CC BY (Waller JLPT lists), CC BY-SA 4.0 (krmanik HSK lists), MIT (drkameleon HSK 2.0 levels); HSK syllabus facts | JLPT N5-N1 kanji (Waller); HSK characters and words per level: the 2025 syllabus (with pinyin), the 2021 standard (人 -> 入 fixed at level 2) and HSK 2.0. | 254 KB |
| ja-zh-compare.json.gz | CC BY 4.0 (StrokeFox) | Our comparison of KanjiVG and Make Me a Hanzi for every shared code point: stroke counts, matched order, strokes drawn the other way. | 32 KB |
| ko-jamo-strokes.json.gz | CC BY 4.0 (StrokeFox) | Our stroke paths for the 40 hangul letters, in the KanjiVG format. | 2 KB |
License manifest
The manifest lists every input file with its version and sha256 and what we built from it: LICENSES.md. Full license texts are in /data/licenses/ and on Credits.
Using the files
Keep the license of each file: KanjiVG adaptations stay CC BY-SA, modified Make Me a Hanzi graphics stay under the Arphic Public License with their notice of changes, the dictionary extracts stay CC BY-SA 4.0. Our own data, the hangul strokes, kana origins table and the Japanese and Chinese comparison, is CC BY 4.0: credit StrokeFox with a link.