Datasets:

---

docling-project

/

regression-dataset-for-docling-parse

Like 3

Follow

Docling 1.15k

Groundtruth for docling-project/docling-parse#349 (fold curly double quotes to ", not ').

Regenerated with pytest --update-groundtruth tests/test_regression_parse.py on that branch (commit 4eaebf4). 148 pages in 77 documents change, each only by ' becoming " where the source has U+201C/U+201D/U+201E/U+201F; every other selected page regenerates byte-identical to the current dataset, so only those pages' files are touched. The page list is in the PR comment: https://github.com/docling-project/docling-parse/pull/349#issuecomment-5635631367

Note: 5462902444445925132-4.pdf pages 1, 3, 4, 5 are also touched by the groundtruth for docling-parse#322; whichever lands second needs those four pages regenerated once more on a build carrying both fixes.

Superseded by #6, regenerated on dataset main 65fd81c7 (this one was generated on 571f34b1 and would revert the #351 bbox precision). Closing.