Citation and source policy.

• Indexed scholarly literature: PubMed/PMC, Crossref, OpenAlex, Semantic Scholar, and related bibliographic metadata.
• Peer-reviewed journals, preprints where clearly labeled, meta-analyses, randomized trials, observational studies, and mechanistic preclinical work — ranked by relevance and design quality.
• Regulatory or monograph references when present in product metadata (not a substitute for clinical consultation).
• We do not treat influencer posts, anonymous forums, or brand blogs as primary clinical evidence.

Research candidates are gathered from product ingredients, claimed benefits, category context, and controlled query expansion.

Duplicates are collapsed by DOI, PMID, and normalized title where possible before curation.

• Papers must relate to the product, its ingredients, or a clearly labeled mechanism pathway — not merely the general word “health.”
• Indirect or population-level evidence is labeled as such; we do not imply product-specific RCT proof without product-relevant outcomes.
• When evidence is thin, listings show research-gap language instead of overstating certainty.
• Retracted or severely disputed papers should be deprioritized or removed after admin review.

• Summaries are generated only from supplied candidate metadata and abstracts available to the pipeline.
• Models are instructed not to invent citations, external facts, or clinical recommendations.
• Each summary should map claims to specific papers shoppers can open.
• AI output is moderated; admins can reject or edit summaries that overreach.

Study type badges (e.g., meta-analysis, RCT, observational, preclinical) help you judge strength of design — not guaranteed real-world benefit.

Directness labels indicate how closely evidence maps to the product or ingredient as sold.

Scores and citation sets can change when new literature is indexed, product data changes, or human reviewers correct errors.

Report citation errors with the product URL and paper identifier so we can re-curate.