医学研究文献综述工作流

references/search_and_citation.md

Use database_strategies.md for current request syntax, authentication, response shapes, and paging limits. Commands below are illustrative discovery recipes, not executed searches or complete corpora.

Supplementary web search

Parallel CLI 0.9.3 supports the flags below. Authenticate with parallel-cli auth or PARALLEL_API_KEY; installation and authentication are described in the official CLI guide. It is optional for the review workflow and may incur provider charges.

mkdir -p sources
parallel-cli search "CRISPR sickle cell clinical studies" \
  -q 'CRISPR sickle cell clinical trial' --json --max-results 10 \
  --excerpt-max-chars-total 27000 \
  --include-domains 'pubmed.ncbi.nlm.nih.gov,biorxiv.org,medrxiv.org' \
  -o sources/scoping-web.json

parallel-cli extract 'https://arxiv.org/abs/1706.03762' \
  --full-content --json -o sources/paper-extract.json

--max-results is a cap on ranked returned hits, not the database total. --full-content requests a fuller extraction; unavailable or truncated content still requires retrieval from the original source. Save the URL, extraction status, date, and exact text supporting each inference. Never report a title, web excerpt, or machine summary as if the full paper was assessed.

Citation chaining

• Forward: use Google Scholar Cited by or documented Semantic Scholar / OpenAlex citation queries. Search-engine queries such as "papers citing ..." are supplementary leads, not complete forward-citation exports.
• Backward: inspect reference lists of eligible reports and relevant reviews; retrieve candidate records and apply the same eligibility criteria.
• Track seed papers, tools, dates, retrieved counts, and screening decisions. Link preprint, journal, protocol, and follow-up reports to a study ID before synthesis. Citation-count thresholds and journal prestige are not eligibility rules; recent and negative studies must have an equal chance of inclusion.

DOI checks and bibliographic identity

python scripts/verify_citations.py review.md --output review_citation_report.json

Optional --email supplies a real contact to the Crossref polite pool. The helper extracts unique DOI candidates, URL-encodes them, and checks GET https://doi.org/api/handles/{doi}. Registration requires a successful Handle responseCode=1 and matching identifier, not HTTP 200 alone. It then requests GET https://api.crossref.org/works/{doi} and checks metadata identity. These public singleton reads require no key or pagination.

Read the report's verified as registration confirmed, failed as unresolved/missing/inconclusive, and metadata_unavailable as manual metadata follow-up needed. Transient errors and throttling are not proof of fabrication. Crossref does not cover every registration agency. The script preserves the original Crossref message and full author list; its compatibility formatting methods are previews, not compliant APA or Nature output.

DOI extraction is heuristic: inspect ambiguous trailing punctuation and older suffixes with parentheses. Citations without DOIs require manual verification. Check titles, authors, year, version, correction/retraction notices, and that the source actually supports the cited statement. Registered identifiers and accessible URLs do not establish those properties.

Sources: Handle proxy API, Crossref REST API.

Rendering references

Use complete metadata in BibTeX/CSL JSON and a current, venue-appropriate CSL file. See citation_styles.md. With Pandoc, cite using [@citation_key]; manually typed reference strings are not reformatted by selecting a style.

# Requires your populated review.bib and local styles/apa.csl.
python scripts/generate_pdf.py review.md --bibliography review.bib \
  --csl styles/apa.csl --output review.pdf

No CSL files are bundled or downloaded automatically. A sibling .bib is recognized; otherwise provide --bibliography or Pandoc bibliography metadata. Omitting --csl uses Pandoc's default citation style. --citation-style remains an alias for a local file/name (e.g. apa requires apa.csl).

Inspect the rendered PDF and all warnings, including unresolved keys. Match in-text citations to the bibliography and manually verify formatting against the destination journal's instructions. Pandoc citation documentation.