citation-verification

Verify each citation by finding its DOI, arXiv ID, or publisher page, then confirm the title, first author, year, and venue against CrossRef, arXiv, or Semantic Scholar metadata. Fetch the BibTeX from a programmatic source rather than generating it from memory.

Never trust AI-generated citations directly, since they have roughly a 40% error rate. Verify every citation through web search and canonical scholarly sources during writing, and mark unverifiable entries with a [CITATION NEEDED] placeholder instead of guessing.

CrossRef is most reliable when a DOI exists and supports direct BibTeX retrieval. arXiv works for preprints via arXiv ID. Semantic Scholar covers general title-based search but requires manual confirmation of matches.

No. Google Scholar is useful for manual discovery and fallback lookup, but it is not a canonical verification authority. If only Google Scholar finds an item, mark it as manually verified and confirm the metadata elsewhere before finalizing the BibTeX.

Title must match with minor differences allowed, at least the first author must match, the year may differ by plus or minus one to account for preprints, and the publication venue must be correct. When citing a specific claim, confirm the claim actually appears in the paper.

The Python scripts are reference implementations, not the primary workflow, and require bibtexparser, requests, semanticscholar, and arxiv packages. They are subject to API rate limits and are intended for batch verification or learning the verification logic.