fix(check): a nested line's words stay out of the line around it - #7705

Context: I've been using HyperFrames every day for the past two weeks to make Korean Shorts and long-form videos, and it's been great to work with. This PR is one of a batch of fixes for things I ran into along the way, so you'll see several PRs from me at once — sorry for the pile-up, and thanks for your patience. Each one is reproduced on current main, independent of the others, and can be reviewed, merged or closed on its own.

What

The craft sampler no longer counts a nested reading block's words a second time in the block around it. The same text is no longer reported twice by line_unread, once on the parent and once on the child.

Why

craftTextParts (packages/cli/src/commands/layout-audit.browser.js) gave every text element the innerText of its reading block. innerText includes block-level children, which readingBlock treats as lines of their own. An end card like

<div id="p3-pp0" class="pp">설계자 <div>코드는 코덱스</div></div>

turned into two lines, and both were reported:

line_unread | #p3-pp0                       | The film's last line "설계자 코드는 코덱스" (3 words) …
line_unread | #p3-pp0 > div:nth-of-type(1)  | The film's last line "코드는 코덱스" (2 words) …

The child's two words were counted in both lines. A mid-film title with a nested subtitle <div> gets the same double warning and an over-counted floor. This happens on any flex row too: a flex container's children are block-level, so .hd { display:flex } with <div>!</div>EMERGENCY ALERT<div>now</div> was read as "! EMERGENCY ALERT now".

Related work

• No open issue or PR (searched line_unread, last line, craft).
• fix(cli): count line_unread words from the block's rendered text #6456 introduced block text so that per-word and per-glyph spans count as one line. That reading is unchanged: a block without a nested reading block still uses innerText.

How

• New readingBlockText(block, root): if any child node is a reading block of its own (readingBlock(child) === child, not counting <br>), the block's text is built from its own text nodes and inline children, and each nested block becomes a separator. Otherwise it returns innerText as before.
• Computed once per block per sample (a Map in craftTextParts), where it used to run once per text element. For a line split into one span per glyph, that is one read instead of one per glyph.

Test plan

• New happy-dom test keeps a nested line's words out of the line around it (layout-audit.browser.test.ts) fails on main (the outer line reads "설계자 코드는 코덱스") and passes with the fix. A second test keeps inline word spans as one line.
• layout-audit.browser.test.ts: 222 passed. layout-audit.chromium.test.ts run locally with PUPPETEER_EXECUTABLE_PATH set to chrome-headless-shell: 83 passed.
• Real CLI on an end-card repro (hyperframes check --no-cache --json, card held 1.5 s):
  • main: #p3-pp0 "설계자 코드는 코덱스" (3 words) and #p3-pp0 > div:nth-of-type(1) "코드는 코덱스" (2 words).
  • this branch: #p3-pp0 "설계자" (1 word) and the child "코드는 코덱스" (2 words). No words are counted twice.
• oxlint, oxfmt, fallow audit --base origin/main, typecheck (pre-commit).

Each chunk of a closing card is still reported as its own "film's last line". I am raising that separately, since it is a reporting-shape question.

🤖 Generated with Claude Code