Use replacement characters for invalid UTF-8 after percent decoding while preserving raw-fragment precedence. Keep the valid UTF-8 buffer reuse path and cover incomplete sequences, valid prefixes and literal ID matches.
Apply the existing serialized-state filter before specialized branches, layout wrappers and target output. Cover hidden textarea and MathML payloads while retaining hidden prose, short JSON and overridden display declarations.
Distinguish normal text, inline code and pre blocks when counting readable content. Honor preformatted_code for inline code and keep blank pre content consistent with the Markdown writer, including referenced code that switches output paths.
Pass the active writer link into inline HTML fallback while keeping block output outside Markdown formatting. Keep audio resources inline inside code so HTML reparsing preserves author links and unique fragment IDs.
Keep authored open attributes on dialogs and details so retained elements preserve their visibility state. Test boolean attribute presence, including empty and false-valued attributes.
Separate synthesized input values from adjacent values and labels in structural HTML output. Reparse paired numeric and labeled fields across all output paths.
Queue targets for sections and other omitted table elements at the next converted cell. Preserve trailing empty-group targets after the table without enlarging each cell plan.
Remove redundant vector scans during target collection and retain global HashSet deduplication. Cover wide MathML target sets and duplicate IDs without a wall-clock threshold.
Collect table header IDREFs independently of fragment navigation, and preserve target IDs on header cells. Validate that each headers reference resolves after HTML reparsing, including adapters that report no fragment links.
Track readable content independently of tags and fragment targets in both writers. Propagate child content marks through captures and carry HTML content metadata, so empty wrappers do not suppress accessible names.
Share structural classifications with the HTML fallback so authored blocks retain boundaries. Cover adjacent sections and related semantic blocks, and retain the authored center wrappers in the layout fixture.
Track active HTML links and render nested media destinations as text, matching the Markdown path. Cover video, audio and iframe content in body, GFM and HTML table output.
Use inline span markers for synthesized fragment targets so a target inside an author link does not create nested anchors. Check link ownership after rendering and reparsing output.
Use the current control-value and accessible-name helpers in structural HTML fallback. Preserve MathML wrappers and surrounding whitespace while filtering visual alternatives.
Cover formatted math links with ordinary and nonbreaking spaces in both Markdown and structural HTML output.
Use ICU punctuation and symbol categories with an ASCII fast path. Apply the same non-whitespace, non-punctuation rule to both delimiter boundaries and cover combining marks, format characters, symbols, and adjacent links with rendered round trips.
ldm0
lanyue-llk