A reference is useful only when it actually supports the nearby claim. We prefer the strongest practical source for the question rather than adding citations for appearance.
Source hierarchy
| Source level | Typical use | Examples |
|---|---|---|
| Primary authoritative | Definitions, official guidance, instrument documentation, laws or public-health recommendations | Government agencies, professional bodies, official publishers, instrument manuals or source documentation |
| Evidence synthesis | Broad conclusions about associations, outcomes or treatment evidence | Systematic reviews, meta-analyses, major evidence reviews |
| Peer-reviewed primary research | Specific studies, mechanisms, validation findings and emerging questions | Peer-reviewed journal articles with a population and method relevant to the claim |
| High-quality institutional explanation | Plain-language context when the underlying evidence is established | University medical centers, established nonprofit health organizations and specialist institutions |
Claim-to-source rules
- Material factual claims should be traceable to a source close enough that a reader can tell what it supports.
- A source about one population is not presented as direct evidence for a different population without explaining the limitation.
- Correlation is not rewritten as causation.
- A single study is not presented as a settled consensus when the wider literature is mixed.
- Secondary explainers do not replace an available primary or official source for a central claim.
Clinical and safety claims
Content about symptoms, diagnosis, treatment, self-harm, crisis, medication, screening thresholds or other high-impact health decisions receives the highest sourcing threshold. Where official guidance or validated instrument documentation exists, we prioritize it and make clear when a tool is a screen rather than a diagnosis.
Clinical acquisition pages remain subject to the separate clinical review gate. Evidence research alone does not equal qualified clinical review.
Statistics, scores and numerical claims
Numbers need a source that supports the number, population, timeframe and measurement being described. We do not manufacture precision from a broad finding. When a statistic is likely to change, we record the source date and recheck it when the page is materially updated.
Uncertainty, disagreement and limitations
When high-quality sources disagree, the page should explain the disagreement rather than force a false single answer. We distinguish established findings, reasonable interpretation and speculation. Limitations belong near the claim they constrain, not hidden in a generic disclaimer at the bottom.
What does not count as evidence by itself
- AI-generated text or an AI summary
- Search-result snippets without checking the underlying source
- Unverified social-media posts or forum comments
- Marketing copy from a company with a direct interest in the claim
- A citation that discusses the topic but does not actually support the statement being made
Freshness and source rechecking
We prefer a current authoritative source when the fact can change, while keeping older foundational research when it remains scientifically relevant. A newer publication is not automatically better simply because it is newer. Meaningful source changes trigger a real content review and, when warranted, a new modified date.