Bottom Line
The explanation half has a coverage problem that works directly against the stated goal. The collection is dense in the long tail (〜べからず, 〜んばかり, 〜甲斐がある, 〜ではあるまいし) and empty at the trunk. There is no page for the て-form, for negative ない, for past た, for the potential, for conditionals, or for sentence-final particles. Those forms appear on nearly every page of every manga, including in this site's own example dialogue. A learner following the grammar index cannot assemble a reading foundation from it; they can only look up patterns they have already met.
Second, quality is split sharply by page age. Pages written in 2026 are strong: formation tables, usage notes, a prose body, and cross-links to confusable neighbors. Pages from 2018 to 2024 are frequently one sentence of frontmatter. Half the live collection (49 of 93) has no formation table and no body text.
Weaknesses
P0 The high-frequency core is missing while rare patterns are covered
The 93 live pages contain no dedicated explanation for any of these: the て-form, negative ない, past た, the potential (読める / ことができる), the volitional, the plain imperative, the causative, the basic conditionals (たら / ば / なら), そうだ, ようだ, たり, ても, のに, もん, わけ, ところ, or sentence-final particles (ね / よ / さ / ぞ / ぜ / かな).
The collection does cover 〜べからず, 〜んばかり, 〜甲斐がある, 〜ではあるまいし, いざ〜となると, 〜つつも, and 〜からこそ. Those are worth having, but a learner will meet ない a hundred times before meeting べからず once.
The clearest evidence is the site's own example dialogue. I extracted all 424 Japanese dialogue lines from the published examples and counted what appears in them:
| Feature appearing in the site's own panel dialogue | Lines | Share | Has a page? |
|---|
| て-form somewhere in the line | 146 | 34.4% | No |
| Negation (ない / なかった / ねえ) | 91 | 21.5% | No |
| Sentence-final ね / よ / さ / わ / ぞ / ぜ / かな | 88 | 20.8% | No |
| Explanatory んだ / のだ / の？ | 61 | 14.4% | Yes, 〜のだ |
| から | 45 | 10.6% | Partly, 〜からこそ only |
| Conditional たら / れば / なら | 23 | 5.4% | No |
| ても / でも | 21 | 5.0% | Partly, 〜てもいい only |
| けど / が contrast | 16 | 3.8% | Yes, 〜が〜 |
| ようだ / みたい | 15 | 3.5% | Partly, 〜みたい only |
| Potential or causative morphology | 13 | 3.1% | Partly, 〜れる receptive only |
These forms already turn up inside the site's Key Points, explained ad hoc where they happen to appear. Call me Teacher with と呼べ explains the plain imperative in a bullet because the panel needed it. That is a symptom: the same explanation is being rewritten per-panel instead of living on a page that examples can point to.
Recommendation: Treat this as the main body of remaining work. _docs/grammar-page-recommendations.md already identifies much of it and grammar.md carries a commented-out TODO list covering causative, そうだ, ようだ, potential, たがる, まい, tense, copula, and the て-form. The analysis is done; the pages are not written. Sentence-final particles and the negative are not on either list and should be added, since they are the highest-frequency items with zero coverage.
Method: dialogue lines extracted from all 260 published example pages, furigana notation and <mark> tags stripped, then pattern-matched. Counts are line-level, so a line containing two features is counted in both rows.
P1 Half the live pages are a single summary sentence
68 of 93 live pages have no formation: block, and 49 have neither a formation block nor any body prose. The whole page is a summary line. Trying it out with 〜てみる is the clearest case: one sentence of frontmatter, no formation, no usage note, no body, no related links, despite having five manga examples attached to it.
The split tracks page age almost exactly:
| Year written | Pages | With Formation | With body prose | With related links |
|---|
| 2018–2019 | 32 | 4 | 6 | 12 |
| 2020–2022 | 18 | 2 | 7 | 5 |
| 2023–2024 | 27 | 3 | 4 | 8 |
| 2025 | 6 | 4 | 1 | 2 |
| 2026 | 15 | 14 | 13 | 13 |
Table covers all 98 explanation files including the 5 unpublished ones, so it does not sum to the 93 live figure. "With body prose" counts pages with more than 200 characters of body text.
Why this matters specifically for reading: without a formation table, a learner sees only the dictionary shape of a pattern. In a panel the pattern is conjugated, negated, contracted, or clipped. The 〜すぎる page says the pattern "may be appended to either the stem form of a verb or an adjective or to a noun" in prose, but never shows the three attachments side by side, so a reader meeting 効きすぎ has to work out the stem rule from a sentence. Compare Doing completely or ending up doing with 〜てしまう, which lays out てしまう → ちゃう, でしまう → じゃう, and the past forms in a block. That page teaches recognition; the older ones teach a definition.
Recommendation: Set a floor for live pages of summary + formation + at least one recognition note, and work the backlog in frequency order rather than alphabetically. The 2026 pages are the template.
Evidence: YAML front-matter audit of 98 explanation pages in _grammar. Full list of thin pages in the appendix.
P1 There is no learning path, only a lookup index
grammar.md groups 93 pages under 32 semantic sections and sorts alphabetically by the Japanese pattern. 15 of those sections hold exactly one page. Nothing on the site marks a pattern as foundational or advanced, states a prerequisite, or suggests what to read next. A learner who wants to look up 〜っぱなし is well served. A learner who wants to build up to reading has no route through the material.
The site already has the missing piece and does not connect it: _articles holds twelve reading-skill pieces (Reading readiness, Deciphering vs reading, Tolerating ambiguity, Using context to deal with missing words, Best manga for Japanese beginners). None of the grammar pages link into them, and the articles do not sequence the grammar collection.
Recommendation: Add a frequency or stage field to explanation front matter and render an ordered "core patterns first" route on the grammar index, with the reading articles as the connective tissue. Even a hand-curated list of thirty patterns in reading order, published as an article, would close most of this gap without touching the collection.
P2 The example corpus is narrow in register
107 of 260 examples (41%) come from ご注文はうさぎですか？. The top three series account for 55%. Both dominant sources are gentle slice-of-life titles with young female casts, and the register statistics of the extracted dialogue show the consequence:
| Register marker | Lines | Share of 424 |
|---|
| Polite ます / です | 39 | 9.2% |
| ない → ねえ style reduction | 14 | 3.3% |
| ぜ / ぞ endings | 11 | 2.6% |
| Feminine わよ / かしら / のよ | 8 | 1.9% |
| Rough masculine (俺 / てめえ / 貴様 / やがる) | 7 | 1.7% |
| Keigo (ござ / いたし / いらっしゃ / おります) | 1 | 0.2% |
| Regional dialect (へん / やん / のう / じゃろ / だべ) | 1 | 0.2% |
A learner who works through the whole collection has seen rough masculine speech seven times and keigo once. Shonen and seinen dialogue, villain speech, samurai and elder speech, and the Kansai-speaking character who appears in a large share of manga are all effectively untaught. 名探偵コナン at 21 examples is the main counterweight.
Recommendation: When choosing panels for new pages, weight toward under-covered registers rather than toward whichever series is at hand. Register belongs in the Key Points too: a line ending in 〜だぜ is a characterization signal, and readers need that read as much as the grammar.
P2 Nothing covers the non-grammatical text on the page
A search across _grammar and _articles found zero pages mentioning onomatopoeia, sound effects, 擬音語, or 擬態語, and none discussing vertical reading order, hand-lettered asides, or the interpretation of … and ーー as pacing. Sound effects are a large fraction of the text on a typical manga page and are frequently the part a learner cannot even look up, since they are rarely in dictionaries and often stylized beyond recognition. Ellipsis already appears in 19.3% of the site's own dialogue lines and long dashes in 2.8%, both unexplained.
Recommendation: This is arguably article material rather than grammar material, but it is squarely inside the stated goal. A page on reading sound effects, and a note in the reading articles on ellipsis and dash as pause and trailing-off, would cover a real blocker cheaply.
P2 Contractions are taught one at a time instead of as a system
The seven Colloquial and Casual Speech pages each document a single contraction. There is no page stating the general processes that produce them: ら-row syllables collapsing to ん, 〜ている reducing to 〜てる, の becoming ん before the copula, 〜てしまう becoming 〜ちゃう. A learner who has read the 〜んない page still cannot decode わかんねー or やってらんない unless those specific forms happen to be covered.
Recommendation: Add one overview page that states the processes and links to the instance pages as worked examples. This converts seven lookups into one transferable skill, and it is high-value per page written for casual manga dialogue.
P2 Vocabulary support on example pages is stubbed out
255 of 261 example files carry todo: add vocabulary, and exactly one has a populated vocabulary: field, despite the field being documented in _docs/example-page-structure.md with guidance to select 3–5 key words. For a reader working through a panel, an unknown word blocks comprehension just as hard as an unknown pattern. The Key Points partly compensate by glossing words that bear on the target grammar, but that is incidental coverage.
Recommendation: Either populate the field or drop it from the documented contract. 255 identical TODO markers suggest a decision was deferred rather than made. If the intent is that the separate _vocabulary collection and words: index carry this load, say so in the docs and remove the TODOs.
P3 Three well-written core pages are finished but unpublished
Marking the topic with は and も (1,720 characters of body prose, formation table, contrastive-は section) and Marking the direct object with を (1,529 characters) both sit behind published: false. Marking the subject with が is live but is the only live page with no manga example attached. These are the three most foundational pages in the collection and the は page in particular is among the best-written.
The apparent blocker is the missing panel examples, which is a reasonable standard. But an unpublished particle page helps nobody, while a published page with textbook example sentences and a note that manga examples are coming helps a lot.
Recommendation: Publish these three with their existing constructed examples, and add panels as they turn up. Alternatively, source panels for them first as a small, high-return task.