# CenjiReader browser edition: third-party notices

Original browser-edition code is MIT. The previous desktop runtime is archived locally and is not deployed. No Python mobi/KindleUnpack or GPL runtime is included in this website.

| Component | Use | License |
| --- | --- | --- |
| React / React DOM / TanStack Query | Interface | MIT |
| Foliate-js mobi.js, John Factotum | DRM-free MOBI6/KF8/AZW3 parsing | MIT |
| @zip.js/zip.js | Local CBZ/ZIP and backups | BSD-3-Clause |
| PDF.js | Local PDF rasterization | Apache-2.0 |
| ONNX Runtime Web | WebGPU and WASM inference | MIT |
| PaddleOCR / RapidOCR PP-OCRv3 detector | Text detection | Apache-2.0 |
| kha-white/manga-ocr-base; mayocream/manga-ocr-onnx | Japanese recognition | Apache-2.0 |
| Kuromoji | Japanese tokenizer | Apache-2.0 |
| IPADIC | Tokenizer data | NAIST / ICOT notices |
| WanaKana | Romaji conversion | MIT |
| JMdict, Jim Breen and EDRDG | Definitions and readings | CC BY-SA 4.0 |
| idb | IndexedDB wrapper | ISC |
| fflate | Compression | MIT |
| LinkeDOM | Inert ebook markup parsing in a worker | ISC |
| Lucide | Icons | ISC |
| Vite / Workbox / vite-plugin-pwa | Build and offline shell | Individual MIT licenses |

Full runtime dependency licenses and version inventory are generated under public/licenses/dependencies by the resource preparation step. Model SHA-256 values and file sizes are pinned in public/resources/models/manifest.json. The Manga OCR ONNX export is revision 24b12778d85800835e2ca409236de281b8ab7b9f of https://huggingface.co/mayocream/manga-ocr-onnx. The detector is the ch_PP-OCRv3_det_infer ONNX model distributed by RapidOCR. Models are not trained or modified here.

JMdict data copyright belongs to Jim Breen and the Electronic Dictionary Research and Development Group. The 64 derived JSON shards remain CC BY-SA 4.0. See public/licenses/JMdict.txt, https://www.edrdg.org/edrdg/licence.html and https://creativecommons.org/licenses/by-sa/4.0/legalcode. Dictionary exports may carry the same attribution obligations when substantially redistributed.

The public website includes compiled application assets, licensed runtime data and notices. It does not publish a source archive or source maps. The repository must remain private. Logos and synthetic study fixtures are original CenjiReader assets.

JLPT level estimates use Jonathan Waller's Tanos lists (https://www.tanos.co.uk/jlpt/, Creative Commons BY), matched to JMdict entries by stephenmk in https://github.com/stephenmk/yomitan-jlpt-vocab (CC BY-SA 4.0). CenjiReader reduces the pinned dataset to entry-ID and exact spelling/reading level maps, retaining the easier level for duplicate assignments. The adapted data is CC BY-SA 4.0; license text, source CSVs, source revision/hashes and reusable maps are distributed in /licenses/jlpt/. These are community estimates, not an official JLPT syllabus. No endorsement is implied.