Skip to content

Releases

citecheck 0.4.1

Choose a tag to compare

@chenyangsu chenyangsu released this 29 Sep 13:32
  • Install from PyPI. pip install citecheck-refs; the command is still citecheck. MIT license.
  • Unusable input is an error, not "clean". A PDF, an empty bibliography, a .txt that is not RIS, or a folder where nothing is cited now exits 2 with a message. An entry that cannot be parsed gives exit 3.
  • Zotero JSON is refused with a pointer to RIS or BibTeX. Before, CSL-JSON was checked with its authors and year silently dropped.
  • RIS and EndNote. Keys like lecun2015deep instead of record numbers; a DOI in the notes is no longer taken as the reference's own DOI; corporate authors such as "World Health Organization," are read as organisations.
  • Fewer false author flags. Organisation authors ({Gemma Team}, LIGO Scientific Collaboration) no longer get a "truncated author list" CHECK. The family name "Ai", hyphenated given names such as "Le-Minh", "Da Silva, Ana Maria" and Vancouver lists with particles are read correctly.
  • Piped input. citecheck /dev/stdin < refs.bib works on Linux and macOS.

Before release, 0.4.1 was tested by CI on Linux, macOS and Windows with Python 3.8, 3.12 and 3.13, and rerun on 7,644 stored evaluation references; every verdict that changed from 0.4.0 was traced to its cause.

citecheck 0.4.0

Choose a tag to compare

@chenyangsu chenyangsu released this 28 Sep 14:20
  • First names. A cited author whose family name is on the record but whose given names share no initial with it gets CHECK. This catches a real paper attributed to namesakes of its authors.
  • Renamed arXiv papers. An arXiv ID that resolves to a retitled paper by the same authors is compared with the title and authors of version 1.
  • Reference-manager exports. RIS (EndNote, Zotero, Mendeley) and EndNote XML files are read directly, so Word users can check their library.
  • Installable. pyproject.toml makes the script pip-installable; the command is citecheck.
  • Fix. "de Winter, Joost CF" is no longer split into two names.

Rerun on the stored evaluation inputs, 0.4.0 changed no verdict on the HALLMARK test split, flagged the ICLR 2026 desk-rejection reference that attributes a real preprint to namesakes of its authors, and added 3 CHECK verdicts among 2,753 references from arXiv papers.

A journal-name comparison was built and left out: on real bibliographies nearly all of its flags were abbreviations, renamed journals or records of reprints.