[adamjohnwright]

Spec 001, T008 (research R4), the low findings of #107's review. None occurs in release 98's labels; each was reproduced as a failing test first, and each fix has a mutant the tests catch.

• Lines are the letters whose heights overlap, directly or through others, so a dot or an accent met first no longer splits a line ("vif" read its x-height).
• A tied baseline vote goes to the lower bottom, so a mark set high does not win ("5′").
• Holes: only a smaller box within another is a hole; two identical subpaths no longer hide each other.
• The tag scan follows every element, leaves comments and character data out, and reads either quotes. A label is unread under a transform by style, a transformed link, a nested <svg>, or <defs>.
• The viewport writes its scale twice, by attribute and by style to six figures; the two must agree. This PR's first version read the real render as unmeasured, because the tests' viewport lacked the style the render writes. The tests' viewport now has both, and the real render reads 7.18 px as before.
• FR-003 names an ancestor's illustration standing in; R4 notes the reader's remaining limits (x-height-only lines, labels other than PATHWAY TEXT not read).

R4's numbers, recounted with this reader, stand. 156 tests pass against the fixture graph.

🤖 Generated with Claude Code

[adamjohnwright]

Replaced by a new PR: its review found two tests that pass without their own fix (the line grouping and the tie-break); fixed there with the review's low findings.