| English | 中文 |
Define what claim or decision the literature should inform. A bounded topic survey, a systematic review and reproduction of a particular study need different search and inclusion designs; choose the design before interpreting the collected papers.
Record search sources, query strings, dates, language/time limits and inclusion/exclusion rules. Prefer original papers and official supplements for methodological claims. Distinguish preprints, accepted manuscripts, published versions and corrections. A paper’s existence or accessibility does not establish access to its data, code or complete model specification.
Deduplicate by DOI and normalized bibliographic identity while preserving version relationships. Link main text, supplements, data and code as distinct evidence objects. Extract a structured record of question, sample, measurements, design, assumptions, comparison, results and limitations, with page/section references. Separate authors’ conclusions from the reviewer’s inference. Compare studies on equivalent populations, outcomes, units and study designs before combining estimates.
Assess selection/publication bias, source coverage, missing attachments, reporting completeness and study-specific validity threats. Search snippets and generated summaries are discovery aids rather than substitutes for the relevant full-text passage. A methodological detail absent from accessible material remains unresolved. Keep inclusion/exclusion decisions and conflicting results visible; do not infer consensus from the number of downloaded files.
State what the evidence supports, for which population and under which assumptions. Distinguish an association, a predictive result and a causally identified effect. Literature synthesis motivates hypotheses and model choices; it does not validate a new dataset or reproduce a computation by itself. Retrieval, archive and import checks are described in the literature workflow skill.