The markup is yours

Varar executes prose. That only stays trustworthy if the prose the matcher sees is exactly the prose you wrote — so Varar follows one rule:

Format plugins own block structure. Nobody touches inline text.

Inline text is explicit

Step matching runs against the raw characters of each sentence. Emphasis, bold, links, inline code — none of it is stripped, normalized or rewritten. Maya borrowed *Emma* contains two asterisks, and only an expression that accounts for them will match.

That sounds stricter than it is. In practice a marked-up run is almost always data — a book title, a product name — and data is what parameter types are for. Emphasis is common enough that it ships built-in as {emph}:

Maya borrowed {emph}, due back on {date}.

Maya borrowed *Emma* matches, and your handler receives Emma — the markers stripped, the value intact. Any other notation that is really data — a hashtag, a wiki link, a domain code — is a custom parameter away. Put the markers in the regexp and a capture group around the part that is the value:

tag: {
  regexp: /#(\w+)/,   // #urgent → urgent
}

The first capture group is what your handler receives — and the exact span an editor highlights — so the # outside it stays notation with no parse needed to strip it. (Leave the capture group out and the whole match is the value, markers and all; that is when you reach for parse.) The markers are notation, no different from the £ in £2.50: the pattern takes them apart, format puts them back, and neither Varar’s core nor your handlers ever see markup they didn’t ask for.

An earlier design stripped emphasis before matching, so *Emma* invisibly became Emma. It felt convenient and read as magic: expressions matched text that wasn’t in the document, and adding emphasis to a word silently changed nothing — until it did. Explicit turned out to be better: what you see in the document is what the matcher sees, and a styling change to a matched sentence fails loudly instead of drifting quietly.

Two practical consequences:

• Styling matched sentences is significant. Emphasizing a word inside a matched sentence changes its text, and the match fails visibly. Narration — the prose around your matched sentences — can use any markup it likes.
• Keep links and inline code out of matched sentences. In narration they’re fine; in a matched sentence the URL is just more characters.

Formats are block structure

What does Markdown actually contribute to an oath? Sections (# headings become example boundaries and describe scopes), prose blocks (paragraphs, list items, blockquotes), tables, and fenced blocks (doc strings and error fences). All of it is block structure. None of it is inline.

That boundary is what keeps the format separable from the matching. The scanner recognises Markdown’s block structure — headings, paragraphs, list items, blockquotes, tables, fences — and hands the matcher only the text inside each block. A block marker like a list bullet or a blockquote’s > prefix belongs to the format, so it never reaches the matcher; everything after it is your text, untouched.

Nothing about sentences, matching or comparison knows Markdown exists. A different block syntax — AsciiDoc’s == headings, |=== tables, ---- blocks — would be a different scanner in front of the same matcher, without changing how a single expression matches; and since {emph} already accepts underscore emphasis (_Emma_), AsciiDoc’s emphasis notation keeps working with no change at all.