Small quality checker for Docling JSON output — feedback welcome

[MMoney1988]

Hi,

I have been using Docling for PDF parsing and built a small downstream checker for the JSON output.

The idea is simple: Docling does the parsing; this small Python project checks the parsed JSON before I use it for downstream steps like Markdown export, manual review, or retrieval-oriented downstream preparation.

Repo:

https://github.com/MMoney1988/pdf-quality-report

Right now it checks things like:

• whether expected fields are present, such as id, type, page_number, and bbox
• whether provenance is still available, including Docling metadata and source PDF context
• whether bounding boxes look structurally valid
• simple content/noise heuristics
• empty, very short, or repeated text blocks

It produces:

• normalized_blocks.json with provenance preserved
• a Markdown quality report with a plain-language interpretation
• a clean Markdown export with source-reference comments

What it does not do:

• no OCR
• no table reconstruction
• no RAG system
• no claim that the extraction is production-ready
• no replacement for Docling

It is meant as a small review layer after Docling, not as a parser replacement.

I would be curious how other Docling users handle this step:

• Do you currently inspect Docling JSON manually, automatically, or not at all?
• Would checks like provenance, bbox sanity, repeated fragments, or content/noise checks be useful in your workflow?
• Are there Docling-specific output patterns that are worth checking more carefully?

Any practical feedback is welcome.