check_spell and check_magic_item compare each mechanical frontmatter field against its bold field line in the corpus body, but the regex that captured a line's value stopped at the first '*'. Shield's Casting Time value legitimately contains an italicized spell name mid-sentence ("...targeted by the *Magic Missile*spell"), so the old regex captured only the text before that asterisk. That forced the corpus's shield.md frontmatter to carry the same truncated value just to pass the check. field_lines now finds only each label's own markup, then computes the value's span in code: from the end of that match to the start of the next field label sharing the line, or the next newline, whichever comes first. Any '*' inside the span turns into a space rather than being deleted outright, since the vendored source sometimes runs an italicized word straight into the next with no space in between, and the resulting whitespace collapses to single spaces. Running verify against the real corpus now correctly flags rules/srd-5.2/spells/shield.md: its casting_time frontmatter value is still the old truncated text, so it no longer matches the full body field line. That file is being tracked separately and was not touched here. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>