From <article-meta>, the JATS backend reads only the title, the authors with their affiliations, and the abstract (_parse_metadata). The copyright and license statements in <permissions> and the funding statements in <funding-group> are never visited, so their text and their links are missing from the document.

Issue resolved by this Pull Request:
Resolves #4794

Where the statements go

The issue asks where these statements belong. This PR puts them in the furniture content layer:

• docling-core defines the body layer as the main content and the furniture layer for content around it, such as page headers and footers. A license or a funding statement describes the article, but it is not part of its reading text.
• Other backends already use furniture for this kind of content. The HTML backend adds the page <title> as furniture "since it is part of the document metadata", and the content of <footer>, where sites put copyright and license notices, also goes to furniture. The Word and Pages backends put headers, footers and footnotes there, and the ODF backend its footnotes; the iWork backend gives the reason: "available to callers that ask for it but stay out of the reading order by default".
• Improve JATS parsing with nested lists, inline formulas, ordered lists, content layers #1067 lists "distribute content in body or furniture content layers" as a goal for this backend.
• The default Markdown, text and chunking output of an article does not change. The statements are in export_to_dict(), in doc.texts and in any export that includes ContentLayer.FURNITURE.

If you prefer them in the body, for example after the abstract under their own headings, that is a small change and I can update the PR.

Change

• _parse_statements collects permissions/copyright-statement, permissions/license/license-p and funding-group/funding-statement from the article metadata in source order. Like abstract paragraphs, they are kept as nodes and walked at emit time.
• _add_statements walks each statement with _walk_linear, so <ext-link> targets and styling are kept as in body paragraphs, and emits it under the title with _emit_inline in the furniture layer. A statement with several runs gets an inline group, also in the furniture layer.
• _emit_inline takes an optional content_layer. The existing callers do not pass it and do not change.
• The statements are added after the back matter and the sub-articles. As furniture, their position matters only when a caller includes that layer, and then they follow the article as article information. Adding them last also keeps every existing item at its index, so the reference data only grows.

Scope: I included <copyright-statement> because it is the other statement in <permissions> and a license is attributed with it. The structured funding data (<award-group>, <funding-source>, <award-id>), the xlink:href of <license> and <ali:license_ref> are not converted. They are identifiers rather than statements, and DoclingDocument has no document-level metadata field for them. In all four fixtures with a license, the license URL is also the target of an <ext-link> in the license text, so the URL is kept. One known limit: block children of <license-p> (a list, formula, figure or table) are still emitted as body content; none of the fixtures has one.

Behavior on the input from the issue:

• Before (main): CHARLIE-LICENCE, CHARLIE-FUNDING and the license URL are not in export_to_dict().
• After: doc.texts gets CHARLIE-LICENCE terms of the, licence (hyperlink https://example.org/c3-licence), . and CHARLIE-FUNDING statement., all in the furniture layer. export_to_markdown() is unchanged. With included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE}, the two statements follow the body.

Reference data

Only the .json files change, and only by additions (503 lines added, none removed). The .md and .itxt files do not change, which confirms that the body is the same.

• elife-56337: the license (with its CC0 link) and the funding statement.
• pntd.0008301: the license (CC0 link) and the funding statement (link to usaid.gov).
• pone.0234687: the license (CC0 link) and the funding statement (links to the three funders).
• ptag100: the copyright statement and the two license paragraphs (CC BY link, and SCOAP<sup>3</sup> as a superscript run). Its <funding-group> has only an <award-group>, so no funding statement.
• pmc2231364: no change (no <permissions> or <funding-group>).

This PR and #4813 regenerate the same JSON reference files (pntd.0008301, pone.0234687, ptag100). Whichever merges second needs a rebase and regenerated reference data. I will do that.

Checklist:

• Documentation has been updated, if necessary.
• Examples have been added, if necessary.
• Tests have been added, if necessary.

Review evidence:

New tests in tests/test_backend_jats_metadata.py. tests/test_backend_jats.py has 1494 lines and the max-lines hook allows 1500, so the tests are in a new module that reuses get_converter, as tests/test_backend_pptx_charts.py does.

• test_jats_license_and_funding_statements_are_furniture: the input from the issue. Checks the furniture runs and the license link, that the default Markdown has the abstract but not the statements, and that the statements follow the body when furniture is included.
• test_jats_copyright_and_each_license_paragraph_are_kept_in_order: a copyright statement, two license paragraphs (one with <sup>) and two funding statements, each kept as its own item in source order; <copyright-year> and <award-group> are not converted.

Both fail on main (f690a52) and pass with this change.

Commands and results (Windows 11, Python 3.10.20, docling-core 2.100.0, environment from uv sync --frozen --group dev --all-packages --no-group docs --no-group examples --extra format-xml-jats):

Not run: the full test suite and the ML suites. On Windows, the hooks that call python3 ran through a python3 alias of the project interpreter.