data/jmdict/README.md

# data/jmdict - the compact JMdict projection, pinned

Generated by `scripts/build_jmdict.py`. **Do not hand-edit or re-compress this file**:
`tests/test_data_assets.py` compares it with the SHA-256 below and pins the entry count,
so any change fails the quick loop until this file is regenerated. The file is a Git
LFS object (`*.gz` in `.gitattributes`); if it starts with `version https://git-lfs`
instead of the gzip magic, run `git lfs pull`.

**Resolved:** 2026-09-06  
**Pins:** scriptin/jmdict-simplified release `3.6.2+20260831182826` (JSON `version` `3.6.2`, `dictDate` `2026-08-31`), 218,672 words  
**Input:** `https://github.com/scriptin/jmdict-simplified/releases/download/3.6.2%2B20260831182826/jmdict-eng-3.6.2%2B20260831182826.json.tgz`  
**Input tgz:** 11,510,336 bytes, SHA-256 `3c842741e2c4f1b780ad4aff6833df308866a8b7dfb9ee59a085e23dd201a65f`

Regenerate (from the repo root; re-running is idempotent - byte-identical output, the
gzip header carries no timestamp or name):

```
.venv/Scripts/python.exe scripts/build_jmdict.py
```

## Files

| File | Bytes | SHA256 | Source |
|---|---|---|---|
| `jmdict-compact.json.gz` | 7,662,785 | `073ee1b823485b1e08eb99d7e088d74c46b01470297e20ccf289e4cbac3ceeab` | projection of the tgz above |

Raw JSON 21,751,814 bytes -> gzip level 9 7,662,785 bytes. Loading is one
`gzip.open` + `json.load` (~0.6-2 s depending on the machine; the build prints the
measured time and `tests/test_data_assets.py`'s fixture prints it again) - kept out of
this file so a re-run stays byte-identical.

## Compact record schema

```
{"meta": {"source", "version", "jmdictVersion", "dictDate", "entries", "licence"},
 "entries": [[id, kanji_texts, kana_texts, senses, common], ...]}

id          int    JMdict entry sequence number (the `jmdict_seq` of data/jlpt/*.csv)
kanji_texts [str]  up to 3 kanji forms, file order (empty for kana-only words)
kana_texts  [str]  up to 3 kana forms, file order
senses      [[str]] up to 3 senses x up to 3 English glosses each;
                   senses with no English gloss are dropped
common      0|1    1 if any kanji or kana form is marked common
```

Entries are in JMdict file order; the index in `entries` is the order the ranked lookup
(plan 02-03) tie-breaks on. Counts in this build: 218,672 entries, 22,637 common, 41,216 kana-only, 0 with no English sense.

## Licence

Dictionary data: **JMdict** - Copyright (c) James William Breen and The Electronic Dictionary Research and Development Group, used under the Creative Commons Attribution-ShareAlike Licence (V4.0). https://www.edrdg.org/wiki/index.php/JMdict-EDICT_Dictionary_Project . https://www.edrdg.org/edrdg/licence.html

JSON conversion by scriptin/jmdict-simplified, release 3.6.2+20260831182826; its derived files carry the EDRDG licence. The compact projection in this directory is a derivative of JMdict and is itself CC BY-SA 4.0.

`LICENSES.md` at the repo root is the project-wide record and the page's credits line
carries the attribution (plan 02-11).