Confirmed Findings
P1 Fifteen reading-article images have no alternative text
The reading articles use instructional images, but 15 images ship without an alt attribute. The affected pages include Using context to deal with missing words (8 images), Reading readiness (5), Deciphering vs reading (1), and Tolerating ambiguity (1).
Recommendation: Give each image an alt description that states the instructional information it contributes. Mark decorative images with an empty alt attribute only when surrounding text already conveys the same information.
Evidence: rendered HTML audit of 1,820 images. Manga panels already receive useful generated alt text; this gap sits in article body content.
P1 Lesson structure is uneven across the catalog
Twelve published grammar or vocabulary explanations have no summary field. Seventy-five of 260 published grammar examples have no Key Points, while every published kanji and idiom example has them. Learners therefore receive a different amount of instructional guidance based on which topic they happen to open.
Recommendation: Make a concise summary mandatory for every explanation and use a minimum Key Points standard for every example. Start with the 75 grammar examples, then add comparison, restriction, and context notes where a literal translation would mislead.
Evidence: front-matter audit. Missing explanation summaries include _grammar/Time and Sequence/〜ながら/Concurrent events with 〜ながら.md and _vocabulary/Training with 稽古/Training with 稽古.md.
P2 The catalog has no learner-facing progression model
Grammar, vocabulary, kanji, and idiom pages contain no JLPT level, difficulty, prerequisite, frequency, or ordering metadata. The grammar index groups lessons by linguistic category, and the kanji index is a long alphabetical list. Neither tells a beginner what to study first or what they should know before opening a page.
Recommendation: Add a small common teaching metadata model: learner level, prerequisites, core versus reference status, and recommended next lessons. Expose it first through a beginner path and a filtered catalog rather than adding another large index.
Evidence: metadata inventory found zero published uses of jlpt, level, prerequisite(s), difficulty, frequency, or order.
P2 Thirty-two published teaching pages have no linked public manga example
The site promises examples from manga, but 32 of 199 published teaching pages do not link to a published example. One vocabulary page and 31 kanji pages fall into this group. Their pages can still contain useful prose or word sections, but they break the site's core expectation.
Recommendation: Flag these pages in the authoring queue and label them clearly in the catalog until an example arrives. Prioritize pages that sit in the proposed beginner path.
Evidence: published explanation and kanji pages were matched to published examples by the existing explanation relationship. Examples include _vocabulary/こそあど/Which with どれ.md and _kanji/Entering and putting in with 入/Entering and putting in with 入.md.
P2 Internal content links rely on legacy redirects and one links to an unpublished page
Fourteen generated internal links do not resolve in the built site. Nine legacy lesson routes rely on NGINX redirects instead of the current canonical route. The related link from Natural result with 〜と points directly to an unpublished page, so it cannot resolve at all.
Recommendation: Update source links to canonical collection URLs, remove or label links to unpublished lessons, and extend the existing lint to fail on generated internal routes that do not resolve without a redirect.
Evidence: rendered-link audit. Redirect-backed examples appear in _grammar/Difficulty/〜がたい/Difficult to do with 〜がたい.md; the unpublished target is _grammar/Condition/When the time comes with いざ〜となると/When the time comes with いざ〜となると.md.