Every borrowed asset has a source.
This edition reproduces permitted modern datasets and fonts. Historical specimen images remain at their source.
Modern data and typography.
| Material | Origin / license | Use in zimai |
|---|---|---|
| Word records | CC-CEDICT / MDBG CC BY-SA 4.0 | 2026-10-12T08:13:53Z snapshot for the 16,001 static word pages. Selected lexical fields adapted, with original abbreviated glosses. Adapted lexical records remain available under the same license. The searchable dictionary uses the complete 13 October 2026 snapshot (125,264 records; Dictionary sources). |
| Handwriting recognition | hanzi_lookup by Gábor L Ugray, a port of Jordan Kiang's HanziLookup; stroke data from Make Me a Hanzi and Arphic fonts GNU LGPL 3.0 · Arphic Public License | Commit 01f90c3ab99a8fadf0696c28e5eb097223c500db, shipped unmodified as a separate WebAssembly module that loads only when the handwriting pad opens; complete source in source-data/hanzi-lookup-01f90c3a/. zimai prepares the strokes and adds CC-CEDICT word context; recognition runs on the reader's device and drawings are not sent anywhere. |
| Stroke outlines and order | Make Me a Hanzi; original Arphic graphics Arphic Public License | Commit bddc96d41bef78427ed0e034e9f7e31d71fd1b92. 32 extracted JSON records; geometry/order unchanged. © 1999 Arphic Technology Co., Ltd. No upstream etymology fields used. |
| Stroke-order medians | Make Me a Hanzi medians by Shaunak Kishore, from Arphic fonts, as redistributed in HanziLookupJS by Gábor L Ugray Arphic Public License | HanziLookupJS commit b21124f771ddd91f131e62aa478bc188f6ec78d3, library/data/x-mmah-medians.js (sha256 8d38024a9789…); vendored unchanged in source-data/hanzi-lookup-js-b21124f7/. 6,963 characters repackaged as one JSON file each with a notice of change; points and stroke order unchanged. The stroke-order player draws them as an even pen line and loads one character's file only when its card comes into view. © 1999 Arphic Technology Co., Ltd. |
| Radical and stroke data | Unicode 17.0.0 Unicode License V3 | Exact fields and per-character locators retained. kTGH membership is not a frequency rank. |
| Interface, English definitions & pinyin | IBM Plex Sans · @ibm/plex-sans 1.1.0 SIL OFL 1.1 | Unmodified, self-hosted complete WOFF2 fonts: Regular, Medium, SemiBold and Italic. Used for dictionary reading and interface text. Font files and checksums. |
| Compact Chinese headwords | Noto Sans SC / TC · Fontsource 5.3.1 SC license · TC license | Original Fontsource Unicode-range WOFF2 subsets at weight 400. Font binaries unchanged. Loaded locally on demand; no external font requests. Large character specimens retain the existing serif fonts. |
| Editorial & retained type assets | Literata / TypeTogether · Source Sans 3 / Adobe Literata license · Source Sans license | Original assets retained from the preceding release. Literata remains in the wordmark and editorial headings. IBM Plex Sans is the primary interface and reading font. Source Sans 3 remains as an on-demand fallback for the four syllabic tone letters Ǹ, ǹ, Ḿ and ḿ. |
| CJK serif type | Noto Serif CJK SC / TC v2.003 SIL OFL 1.1 | © 2017–2024 Adobe. Local WOFF subsets; unchanged glyph outlines. SC for simplified Chinese, TC for traditional Chinese; rare-character system fallback remains. |
| Graphical decompositions | BabelStone IDS · Andrew West | 8,243 exact records from 2025-06-27 / Unicode 16.0; regional qualifiers retained. Publisher permits personal and commercial reuse. Source header and inputs. |
| Optional 3D renderer | Three.js 0.186.1 MIT license | Unmodified self-hosted upstream modules. Loaded only when 3D is enabled; perspective is a visual arrangement, not linguistic evidence. |
| Historical discussions | CUHK and identified MOE entries | Original attributed English summaries with direct links. No grant of rights to source specimen images is assumed. |
Project code and original editorial prose have no blanket open-source license selected yet. That does not override the continuing licenses and required attributions for incorporated datasets. Reuse the adapted lexical data under CC BY-SA 4.0 and respect each separately licensed asset.
Download the structured research data · Make Me a Hanzi provenance · CC-CEDICT snapshot header · Font subset notes
Scholarly datasets.
PHONO-ML readings by Jacques and Delaporte / CRLAO are CC BY 4.0. CJKVI / Kawabata component analyses, their adapted selection and the scholarship importer are GPLv2. Source data, original notices, licenses and modifications are available from the scholarly source register. These terms apply to their respective materials. The Hill–List paper and the complete Baxter–Sagart database are not redistributed.
Editorial process.
zimai’s English explanations, classifications and example sentences were assembled and reviewed through an AI-assisted research process. Cited claims were compared with consulted entries, but no independent expert review is claimed. Corrections and research gaps are part of the record.
Chinese-first typography
Noto Serif SC / TC is used for large character specimens; Noto Sans SC / TC is used for dense headwords and network labels, with separate regional fonts. Noto font binaries are unmodified Fontsource distributions: Serif 5.2.9 (regular), Sans 5.3.1 (regular, medium, semibold). Fonts by the Noto / Source Han contributors, under SIL Open Font License 1.1. Original project · Files and checksums · Serif SC license · Serif TC license · Sans SC license · Sans TC license.
Zimai Rare Han provides additional character and component coverage. These are renamed, partitioned WOFF2 subsets of BabelStone Han 16.0.3 by Andrew West, based on fonts by Arphic Technology. Glyph outlines are unchanged; font identification names and encoded coverage are modified by zimai on 2026-10-14. The original Arphic Public License (1999) remains in force. Original source · Modifications and checksums. These modern glyphs primarily follow PRC conventions and are not historical facsimiles.
IBM Plex Sans remains the English and pronunciation face, with Source Sans 3 covering syllabic tone letters. Fonts are hosted locally; Chinese Unicode subsets load only when their characters are used.