Every character page is generated by our code from open data: no one writes the pages one by one, and no text is copied from other sites. This page explains what the code does, what it checks before anything is published, and what that means for accuracy.
From data to pages
The build reads KanjiVG, Make Me a Hanzi, KANJIDIC2, JMdict, CC-CEDICT and the JLPT and HSK lists, computes what each page needs and writes the pages. The sentences are templates we wrote, with the help of an AI model, that change with the shape of the data: a kanji with a part that is written last gets a different paragraph from a kanji that is its own radical. Numbers in the text are counted, never typed.
When a character gets a page
A character gets its own page only when the data says enough about it: strokes, a reading, a meaning, at least five common words (two for the first two levels of each list, whose most frequent characters are often words on their own) and a JLPT or HSK level. Pages are released in waves, starting with JLPT N5 and N4 and HSK 1 and 2. Every other character is still available in the viewer. The rule and its changes are on the corrections page.
What is checked
Before publishing, automated checks read the finished site: that every page names the sources whose data it shows, that no two pages are near-copies by our similarity check, that no vulgar sense or raw dictionary notation reaches a page, that every claim that Japan and China write a character differently comes from our stroke-by-stroke comparison, that the tracing accepts each correct stroke and rejects it drawn backwards, that the practice PDFs are valid, and that the dictionary data is no more than 35 days old. Each check is itself tested by breaking the site on purpose.
Data dates
KANJIDIC2 version 2026-286 and JMdict of 2026-10-13; CC-CEDICT of 2026-10-13. The EDRDG licence asks sites to refresh its dictionaries at least once a month, and the build refuses to publish older copies.
Limits
Stroke data reflects one tradition each: Japanese school order and mainland Chinese order. Taiwan and Hong Kong teach some characters differently. JLPT and HSK lists here are unofficial copies. If something is wrong, write to hello@strokefox.com and we will fix it and log the fix.