[PierreMesure]

Titles and section headers can store their rich text in an InlineGroup while their own text field is empty. HierarchicalChunker reads only that empty field, so body chunks lose their heading metadata. It also emits the heading's inline children as separate body chunks.

Capture serialized inline heading content when the heading text is empty. Reuse the captured text for ordinary body chunks and for empty-section heading emission. Mark the inline children as visited so they do not become body chunks. Keep the existing text path for plain headings.

Fixes #925.

This addresses the heading metadata blocker in docling-project/docling#4908. Restoring the parent's text in the adapter alone is not sufficient: the current Markdown serializer then duplicates the heading and escapes math underscores. The downstream PR still needs a core release containing this fix and a dependency update.

Validation on Python 3.12.10:

• Core hierarchical chunker and serialization suites: 244 passed, 1 existing expected failure.
• Nine new cases cover rich titles and section headers, formula-only headings, empty sections and sibling headings. The initial eight cases fail without this change.
• Four downstream DOTS/Nemotron Title/Section-header regressions pass with this core change and fail with locked docling-core 2.103.1. They assert body chunk text, heading metadata and exact Markdown export.
• Downstream focused parser and export suites with this core change: 122 passed.
• make validate passes for the changed core files, using the downstream checkout's development environment.

No reference fixtures were regenerated. The full repository suites, hybrid chunker model/tokenizer suites and model inference were not run.

[github-actions]

✅ DCO Check Passed

Thanks @PierreMesure, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🔴 1 of 2 protections blocking · waiting on 👀 reviews

	Protection	Waiting on
🔴	Require two reviewer for test updates	👀 reviews
🟢	Enforce conventional commit	—

🔴 Require two reviewer for test updates

Waiting for

• #approved-reviews-by >= 2

This rule is failing.

When test data is updated, we require two reviewers

• #approved-reviews-by >= 2

---

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[geoHeil]

Reviewed e98e5bde7f14f1ad68749cdbdc783061cbcc12cc. No blocker found.

A title or section header with an empty text and one InlineGroup, FormulaItem or CodeItem child now gets its heading text from that child. The child is marked as visited, so it no longer becomes a separate body chunk. This fixes the heading metadata defect in docling-project/docling#4908.

Validation on Python 3.10:

• tests/test_hierarchical_chunker.py, tests/test_hybrid_chunker.py, tests/test_serialization.py: 286 passed, 1 xfailed.
• With the base chunker, 9 of the 10 new cases fail. The tests detect the defect.
• Extra cases: an inline heading my_var *x* with the formula a_b gives my_var *x* $a_b$ with no escaping. A heading with non-empty text and an inline child keeps the base behavior.

Suggestion, not blocking: an empty heading with one plain TextItem child, or with several direct children, still gives the heading "". Base has the same behavior, and the DOTS and Nemotron output uses an InlineGroup.

CI passes at this head. I did not run the full repository suite.