mainichi 毎日

A kanji a day, stuck on your desktop like a post-it.

Status: the window, the interaction model and the kanji data are in place. 2211 kanji across the five JLPT levels, each with common words per reading, furigana, example sentences, stroke order animation, and text that can be selected and copied.

Requirements

• Python >= 3.10 with tkinter (standard library, no third-party dependency)
  • Debian/Ubuntu: sudo apt install python3-tk
  • Windows / macOS: bundled with the python.org installer
• A font covering Japanese. Without one, kanji show up as empty boxes.
  • Debian/Ubuntu: sudo apt install fonts-noto-cjk
  • Windows 10/11: Yu Gothic / MS Gothic are already installed

Running

python -m mainichi                 # random kanji from N5 (default)
python -m mainichi --level N3      # random kanji from N3
python -m mainichi --kanji 日      # one specific kanji
python -m mainichi --colour blue --size 360x400
python -m mainichi --language fr    # translations in French (en, fr, es)
python -m mainichi --verso sentences # back of the note: example sentences
python -m mainichi --decorated     # keep the title bar, if borderless misbehaves

--level and --kanji are mutually exclusive, as they should be.

Launch it several times for several post-its, or press Ctrl+N on one.

Using a post-it

Action	Result
left click on text	select that word or sentence; drag to extend over several
left click on blank paper	clear the selection
double click	flip between the front (kanji + words) and the back (meaning + vocabulary or sentences)
drag the blank paper	move the note around the desktop
drag the folded corner	resize the note
hover over a word or sentence	a bubble shows it larger, with readable furigana
right click	context menu (on-top, furigana, translation, language, back of the note, colour, next, ...)
left click outside the menu, or Esc	close the context menu
Ctrl+C / Ctrl+A	copy the selection / select everything on this side
Space	flip
Ctrl+N / Ctrl+W / Ctrl+Q	new note / close note / quit

Closing the last post-it quits the application.

The data

Level	Kanji	Words	Extra vocabulary	Sentences	Furigana	On level	fr	es
N5	79	266	790	231	98%	50%	41%	35%
N4	166	472	1660	426	99%	69%	47%	45%
N3	367	1009	3670	872	99%	88%	59%	54%
N2	367	826	3670	690	99%	87%	52%	45%
N1	1232	2299	10840	1558	99%	97%	47%	36%

Each kanji carries its English meanings, its on and kun readings, and up to four common words — one per reading of that kanji, most common first — with the furigana aligned per character, so 日曜日 shows にち above 日, よう above 曜 and び above the last 日.

The back of the note shows either more vocabulary written with the kanji (the default) or example sentences, one per word on the front. The choice is in the context menu under "Back of the note".

The vocabulary side puts each meaning straight after its word, which leaves enough room for two columns even at the smallest note size: eight words fit where four did before. A bigger note widens the columns and adds rows rather than cramming in a third column, which would only shorten the meanings.

The sentences are chosen for the level: short, everyday, and built from kanji the learner already knows. "Furigana" above is the share of sentences where every kanji could be given a reading; "on level" the share using no kanji above the level. N5 is the hard case — with only 79 kanji available, half of its sentences still need one harder character, which is why full furigana matters.

Translations come in English, French and Spanish, for both the sentence translations and the word meanings. English is always present; the other two depend on what the sources cover, and anything missing falls back to English with a small [en] marker rather than silently showing another language.

	French	Spanish
Word meanings, N5/N4	73-81%	82-86%
Word meanings, all levels	71% front / 49% vocabulary	75% / 56%
Sentence translations	41-59%	35-54%

Word meanings come from the full JMdict rather than its English-only export. Coverage is deliberately best where beginners are: among words of comparable usefulness the build prefers the ones that are translated.

Cards and sentences come to 1.6 MiB, plus 545 KiB of stroke outlines. Each is, gzipped JSON, one file per level, loaded lazily: showing an N5 note never reads the 1232 kanji of N1. The application never uses the network.

To regenerate it (needs an internet connection, takes about ten seconds plus downloads):

python tools/build_data.py            # all levels
python tools/build_data.py --levels N5

Data sources and their licences are listed in ATTRIBUTION.md. Note that the generated data is CC BY-SA, while this code is Apache-2.0, and that the JLPT has published no official kanji lists since 2010 — the level assignment is a well established reconstruction, not an official list.

Architecture

mainichi/
├── cli.py          argument parsing, builds an AppConfig
├── config.py       AppConfig (launch) + PostItOptions (per-note, menu toggles)
├── content.py      KanjiCard / Word / Sentence + the CardProvider protocol
├── dataset.py      BundledProvider: reads mainichi/data, deals shuffled decks
├── strokes.py      KanjiVG outlines: SVG paths flattened into polylines
├── data/           N5.json.gz ... N1.json.gz, generated, committed
├── app.py          MainichiApp: owns the Tk root and every open post-it
└── ui/
    ├── postit.py   PostItWindow: borderless window, canvas drawing, input
    ├── ruby.py     furigana layout and line breaking, shared by note and bubble
    ├── bubble.py   the hover popup that magnifies a word
    ├── theme.py    Palette + the six post-it colours
    └── fonts.py    finds a Japanese capable font family per platform
tools/
└── build_data.py   downloads the open data sets and builds mainichi/data

The presentation layer only ever reads a KanjiCard and asks a CardProvider for the next one, so swapping the bundled data for an online lookup would not touch ui/.

Reading the furigana on a small note

A post-it parked in a corner of the screen is often too small for the furigana over its example words. Rest the pointer on a word (or a sentence) and after a moment a bubble pops up beside it showing that text larger, with the reading at a comfortable size — no need to resize the note.

The bubble is a window of its own, so it is not clipped by a small note, and it flips to the other side of the pointer rather than running off the screen. It goes away as soon as the pointer moves off the text or you click. Turn it off with Bubble on hover in the context menu.

It does the same for a meaning that was too long for its column and had to be cut short with an ellipsis: hovering shows the whole of it.

Stroke order

Traces in the context menu writes the kanji out stroke by stroke, in the proper order, over a faint outline of the finished character. The outlines come from KanjiVG and cover all 2211 kanji. Any click, flip or resize stops the animation; choosing Traces again replays it.

The outlines live in their own file (data/strokes.json.gz, 545 KiB) that is only read the first time an animation is played, so notes that never use it pay nothing.

Copying text

Click a word or a sentence to select it, drag to take in several, then Ctrl+C or right click and Copy. Furigana is never copied: the clipboard gets 日本, not にほん, so what you paste elsewhere is the text as written. On X11 the clipboard is owned by the running application, so paste before closing the last note.

Contributing

Bug reports, data corrections and focused patches are welcome — see CONTRIBUTING.md for how to run the tests and regenerate the data. Note that a wrong reading or translation usually needs fixing upstream in JMdict or Tatoeba rather than here.

Not implemented yet

• persisting position, size, colour and options between runs
• 20 of the N1 kanji are used almost only in names and have no common word; those notes show the readings instead