Workflow leaderboard / Research / Reference management

Check reference format and citation round trips

检查参考文献格式与引文往返一致性

Research / 科研 Reference management / 文献管理 curated verified 2026-08-02

Apply a journal CSL style reproducibly, surface unresolved citation keys and verify that in-text citations match the bibliography.

Variants

Free & local / 免费本地 ~25 min setup

Any current laptop; no GPU required.

ComponentRolePriceLink
Pandoc 3.10.1 CSL-based citation rendering and document conversion Free / 免费GPL-2.0 open-source document converterchecked 2026-08-02 https://github.com/jgm/pandoc/releases/tag/3.10.1
Citation Style Language styles 0.2.170 versioned journal citation styles Free / 免费open style repository; individual style metadata states its licensechecked 2026-08-02 https://github.com/citation-style-language/styles/releases/tag/v0.2.170

Render with a pinned style and capture warnings

mkdir -p build
pandoc manuscript.md --citeproc --bibliography references.bib --csl journal.csl --metadata link-citations=true -o build/manuscript.docx 2> build/citation-warnings.txt
test -s build/manuscript.docx
if rg -n 'citation.*not found|could not find|unresolved' build/citation-warnings.txt; then exit 1; fi
sha256sum manuscript.md references.bib journal.csl build/manuscript.docx > build/SHA256SUMS

Reference-format acceptance checklist

[ ] CSL style name and repository revision recorded
[ ] Every citation key resolves
[ ] Every bibliography entry is cited unless intentionally listed
[ ] DOI and URL fields open successfully
[ ] Author order and particles checked on a sample
[ ] Journal title and capitalization match source metadata
[ ] In-text disambiguation checked
[ ] Final document compared with the journal's current author guide
[ ] Bibliography regenerated, never manually reformatted

Known pitfalls

  • A valid CSL render does not prove metadata is correct.
  • Journal author instructions can diverge from the repository style.
  • Manual edits to the rendered bibliography will be lost on rerender.
  • Name particles, institutional authors and translated titles need human sampling.

Evidence