出典Where the data comes from
The flip-cards draw on the same frequency-built lists as Nihongo Dominos, and the stories themselves are built by a comprehension-first content pipeline. Every list traces back to an authoritative open source, and the methodology is described here in plain English so you can check any choice.
語彙 · Goi (vocabulary)
Curated by category — verbs, adjectives, adverbs, nouns, conjunctions, number paradigms — with attention to register balance, then validated entry by entry against JMdict, the open-source Japanese-English dictionary. Pure newspaper-corpus frequency is deliberately avoided: newspapers cover politics and business and almost never use basic everyday verbs like 食べる or 飲む, so raw corpus rank buries the words a learner needs first. Each entry carries an example sentence and a usage note where it helps.
漢字 · Kanji
The pool unions the textbook-standard reference sets: all 1,026 Kyōiku kanji (what a Japanese child learns by age 12) and the full community-canonical JLPT N5–N1 list, plus the remaining non-Jōyō kanji that the app's own vocabulary actually uses — 2,228 kanji in all, so every kanji you can meet in a card has its own card. Within that closed pool, ordering uses a blended frequency score — the average of three independent rank signals across distinct registers: the Mainichi Shimbun newspaper rank (via KANJIDIC2), Japanese Wikipedia, and Aozora Bunko literary prose. Each card's colored-radical glyph and stroke order come from KanjiVG; the mnemonics are written under a strict style guide, using only canonical Kangxi-aligned radical meanings — no invented radical names.
文法 · Bunpou (grammar)
Grammar can't be corpus-extracted — patterns are meaning-bearing structures that don't fall out of a token count, and the JLPT stopped publishing official grammar lists in 2010. So the list was generated by tier under explicit design rules (one specific surface form per tile, variants consolidated) and then audited against the canonical learner references the field already trusts — Bunpro, the Dictionary of Japanese Grammar, Genki, Tobira, Quartet, Tofugu — with a real-text walkthrough to catch gaps.
The stories — a comprehension-first pipeline
The reading content is natural Japanese where, ideally, every content word maps to a flip-tile in the shared data. Stories aren't auto-generated from a word list; they're written to read naturally, then run through a comprehension pipeline that decides, for each word, which tile it links to in context. That last step matters: same-surface homographs (昨日 / きのう as a word vs. its kanji; readings that collide) can't be resolved by string-matching alone, so an AI that understands the whole sentence makes the call, and a final sensei-grade read-through verifies the mappings before a scene ships.
Sources
Every word, kanji, grammar pattern, and example traces back to one of the public sources below — listed with maintainer, license, and what it's used for, so you can verify any choice or rebuild the data yourself.
JMdict — vocabulary dictionary
Validates every Goi entry — each must exist in JMdict, which verifies spellings, readings, and part-of-speech tags. The canonical Japanese-English dictionary used by IMEs and learner tools everywhere.
KANJIDIC2 — kanji dictionary
Supplies kanji readings (kun + on), English meanings, Joyo grade, stroke count, and the embedded Mainichi Shimbun newspaper-frequency field used as one of three ranking signals.
Mainichi Shimbun frequency rank
One of three signals in the blended kanji ranking — the formal-news register. Blended with two complementary corpora rather than used alone.
Japanese Wikipedia — modern signal
Second of three kanji-ranking signals — a modern, broad-topic encyclopedic register sampled via the MediaWiki API. Lifts elementary kanji (市, 県, 学, 山) that newspaper alone under-ranks.
Aozora Bunko — literature signal
Third kanji-ranking signal — public-domain literary prose from canonical 20th-century authors (Sōseki, Akutagawa, Dazai, Miyazawa, and others). Restores daily-life and emotional kanji (顔, 心, 寝, 雨) the formal corpora miss.
KanjiVG — stroke SVGs + decomposition
Powers the colored-radical kanji card — both the stroke geometry for the ▶ animation and the radical decomposition for the tappable chips. The only freely-licensed dataset that provides both.
JLPT level reference (community-canonical)
Assigns each kanji a JLPT level. The JLPT stopped publishing official kanji lists in 2010; this community-compiled list is the de-facto standard the learner ecosystem uses.
Bunpou cross-references (grammar)
Cross-checked against Bunpro, Tofugu, the Dictionary of Japanese Grammar (Makino & Tsutsui), Genki, Tobira, and Quartet — six independent sources across different audiences that converge on a stable working set of patterns.
What's deliberately not used: no proprietary corpora (every signal above is public), no machine-generated content lists shipped unchecked (AI drafts the curation passes, but every entry is validated against a dictionary or audited against the references above), and no invented "WaniKani-style" radical names you'd have to unlearn later.
日本語読書 · NihonGO Dokusho