A constrained bold span closing beside a pass:[] macro is refused because the boundary test reads the unmasked text

What happens

A constrained bold span whose closing mark abuts a pass:[] macro is not read as a span:

Bytes round-trip; the divergence is parse-side (a tree pin, not a format pin). The sibling kinds are already right: _a_pass:[b], `a`pass:[b], #a#pass:[b] and "`a`"pass:[b] all build their span (measured in the #194 change, whose passthrough mask reached them).

Where it is decided

canOpenAt / canCloseAt in src/parse/inline/quote-boundaries.ts (around l.465 and l.503) choose source = seesCurvedRewrite(kind) ? view : text. Every kind but bold reads curved.view, which since #194 is derived from the passthrough-masked text; bold's unconstrained row sorts ahead of the curved rows in QUOTE_SUBS, so seesCurvedRewrite("bold") is false and bold reads the raw fragment, where the p of pass: is a word character that refuses the closing boundary. The Ruby sees a non-word placeholder there (extract_passthroughs runs before the quote pass, substitutors.rb). The inner non-space test at l.455 and l.499 reads the raw fragment for every kind and has the same gap.

Fix shape

Carry the passthrough-masked fragment through the window the way curved.view is carried, and read it in both boundary tests: a masked-fragment field on InlineScan in src/parse/inline/quote-pass.ts (clipped in windowOf), handed down by markFlags in src/parse/inline/rules.ts, read by both boundary functions in quote-boundaries.ts. Tree-only, byte-safe, the same class as #194. Found by the #194 review (2026-09-06).

Start with the boundary tests in src/parse/inline/quote-boundaries.ts and trace InlineScan through windowOf in src/parse/inline/quote-pass.ts and markFlags in src/parse/inline/rules.ts; compare the passthrough handling from #194. Verify that x apass:[b] c produces a bold span while the existing sibling cases and byte output remain unchanged.