Re: [jats-list] [ANN] Feedback sought on a DOCX-to-JATS content-preservation prototype

["Alejandro Federico Moyano afmoyano@xxxxxxxxx" <jats-list-service@xxxxxxxxxxxxxxxxxxxxxx>]

Hi,

Your approach to validating content preservation between source documents and
target JATS XML addresses one of the most critical friction points in academic
publishing.

At Estudio 2A (https://estudio2a.ar/), we work on single-source publishing
workflows and developed an open-source tool called gbpublisher
(https://github.com/albertomoyano/gbpublisher) to handle XML production for
scholarly journals. The project is licensed under Business Source License (BSL
1.1)bit is open-source and completely free for Diamond Open Access journals
(specially designed to empower universities in the Global South, which is why
the UI and documentation are in Spanish), while restricting free commercial
use for publishers that charge APCs. We welcome anyone in the community to
explore the repository and collaborate.

Based on our experience converting author manuscripts to JATS 1.4, here are a
few insights regarding your questions:

1. Which content-preservation checks take the most manual effort?
- Unstructured References (<mixed-citation>): A recurring issue when
converting directly from Word is references ending up as flat <mixed-citation>
elements instead of fully parsed <element-citation> nodes. This directly
interferes with citation tracking and impact metrics. Verifying that authors,
titles, publication years, and DOIs are properly structured requires
significant manual validation.
- Citation & Reference Linkage: Ensuring that every inline citation (<xref
rid="...">) correctly maps to a unique <ref> item in the back matter,
especially when authors use non-standard citation formats or ambiguous
labels.
- Complex Tables & Cell Formats: Validating table cell values, colspans,
rowspans, and inline formatting (sub/superscripts, math symbols inside cells)
without losing structural integrity.
- Special Characters & Entities: Verifying Unicode characters, math symbols,
and custom glyphs across Word styles and XML parsers.
- Footnote/Endnote Placement: Ensuring footnotes attached to headings, table
cells, or inline text remain attached to their exact semantic parent nodes in
JATS.

2. Workflow approach: Database + Markdown -> Multi-format Output
In our experience, relying solely on DOCX files for final generation
introduces noise due to MS Word's proprietary/inconsistent styling. We find it
much more reliable to manage and validate manuscript metadata canonically
using a dedicated database engine.

For the body content, our pipeline normalizes the manuscript into an
intermediate structured Markdown stage (using Pandoc with custom Lua filters
and Saxon XSLT 2.0) before generating canonical JATS XML. From that single
source, our system produces:
- JATS XML tailored for different indexers and citation index harvesters.
- SEO-optimized HTML for web delivery.
- Metadata-enriched PDF/A for long-term preservation.
- Accessible EPUB 3.
- Clean ODT files.

In our view, a truly powerful tool in this space must guarantee complete
content traceability and deterministic rendering.

3. Perspective on AI-based parsing
We notice a growing caution in scholarly publishing regarding AI-driven
parsing solutions, as probabilistic models frequently introduce subtle
hallucinations or silent alterations that degrade publication quality. We
believe tools built on strict, deterministic rules and structural
validationbwhere human editors maintain total, traceable controlboffer a
much more viable long-term strategy for the academic community.

4. What we would need to see before considering a trial:
- High deterministic accuracy on complex inline markup (italics, bold,
mathematical symbols, chemical formulas) and reference parsing without
unverified inferences.
- A clear visual diff/traceability report that non-technical editors can
interpret without reading raw XML.
- Support for CLI/batch integration so it can fit into automated pipelines
(e.g., CI/CD or local publishing scripts).

Best regards,

Alejandro Federico Moyano
Estudio 2A (https://estudio2a.ar/)
https://github.com/albertomoyano/gbpublisher



Enviado desde el correo electrC3nico seguro de Proton Mail.

El viernes, 9 de octubre de 2026 a las 19:14, 0520sdh@xxxxxxxxx
<jats-list-service@xxxxxxxxxxxxxxxxxxxxxx> escribiC3:

> Hello,
>
> I'm seeking feedback from people who check whether content survives
> conversion from an approved DOCX manuscript to JATS.
>
> I have an early prototype that compares a limited set of DOCX/JATS
> structures and flags differences with source and output locations. Its
> scope includes paragraphs, table values, captions, footnotes, and
> citation/reference correspondence.
>
> It has only been tested on synthetic documents. It has not been
> validated on real publication workflows or across Microsoft Word
> versions. I'm exploring whether this could become a useful paid tool;
> it is not a production-ready product.
>
> Would anyone involved in publishing or XML production be interested in
> reviewing the approach? In particular:
> - Which content-preservation checks take the most manual effort?
> - Do your existing tools already cover these checks?
> - What would you need to see before considering a trial?
>
> Please describe workflows without sharing confidential manuscripts or
> personal data.
>
> Thank you.