Bee's Ultimate Grammar Dictionary

13 Grammar Dictionaries in one

1. Download the dictionary ZIP
2. In Yomitan settings, go to Dictionaries → Import
3. Select the downloaded ZIP

You can also browse the grammar library or use the English website. Every dictionary entry links to its page, with all contributing explanations and examples. Search by grammar, reading or meaning, and filter by JLPT level or source.

The English Yomitan edition is separately installable. Monolingual explanations and untranslated examples are translated with GPT-6 Luna and labelled as translations. Source-authored English is retained. The English website includes the original source text for reference.

Every grammar point shows a compact card with progressive disclosure — click to expand per-source details, extra examples, and provenance.

• Single lookup: one dictionary, one card per grammar point — not 13 separate popups
• Merged entries: when multiple sources describe the same grammar point, their contributions are merged into one card with per-source attribution
• Deduplication: the keymap stage aligns entries across sources so you don't see the same point repeated
• Progressive disclosure: compact by default, expand any source's details on demand

python3 -m venv .venv
source .venv/bin/activate
pip install . pytest
make all

The pipeline stages:

1. extract — read each source's locked data into normalized GrammarPoint records
2. keymap — align entries across sources (which rows are the same grammar point?)
3. merge — combine aligned entries into a unified dataset
4. build — render the Yomitan dictionary ZIP
5. validate — verify against Yomitan's JSON schemas

The committed publication snapshot makes both websites and dictionary downloads reproducible without local source decks, network scraping, or model credentials:

make site
python3 -m http.server 8000 --directory build/site

Open http://localhost:8000/ or http://localhost:8000/en/. GitHub Actions builds this snapshot and deploys it to GitHub Pages on main. See the publishing workflow for updating the corpus and translations.

After acquiring all locked sources and running make all, update the publication:

make translate-english  # authenticated Codex CLI; explicitly uses gpt-6-luna
make publish-data      # refuses missing sources or translations
make site

Translation results are cached by source-text hash under data/translations/. The original corpus stays unchanged. A changed source text requires a fresh translation; an incomplete English edition fails the build.

1. Place source data in data/sources/<name>/ with a SOURCE.lock.json
2. Write an extractor in src/bugd/sources/<name>.py (see existing extractors for patterns)
3. Register it with @register_extractor
4. Add to SOURCE_PRECEDENCE in keymap.py and _SOURCE_EXPLANATION_LANG in banks.py
5. Run make all

Dictionary content retains its original licensing from each source. The build pipeline code is MIT.