Start with the question
Before deciding whether evidence is persuasive, write the claim in a form that can be checked: this exact material, used by this route, in this population, changed this measured outcome compared with this comparator. If an article cannot supply those details, its conclusion may be broader than its source.
For example, increased collagen production in a culture dish is not the same question as visible improvement in human skin. Neither result should silently stand in for the other. A mechanism can justify a hypothesis while leaving the clinical question unanswered.
Separate the study models
Use the following distinctions to navigate the Evidence Finder. The labels describe the setting, not a universal ranking of quality.
| Source type | Useful for | Does not establish by itself |
|---|---|---|
| Cell or laboratory study | Mechanisms and analytical observations | A clinical benefit or a human regimen |
| Animal experiment | Responses in the particular model | Matching human exposure or outcomes |
| Uncontrolled human report | Observations that may guide research | Causality or reliable comparative benefit |
| Controlled human study | A defined comparison in a studied population | Every formulation, route or use |
| Product or registry document | Identity, study metadata or official labeling | An independent efficacy result |
Read design before confidence
A human label is not a quality seal. Ask how participants were selected, whether allocation was randomized, who knew the assignment, what comparator was used, and how outcomes were collected. A small retrospective series can be relevant without being definitive.
Keep the number enrolled, the number completing follow-up and the number analyzed separate. Missing follow-up and a change of endpoint can alter the meaning of a result. Look for funding and author relationships as part of the context, not as automatic proof for or against a finding.
A repeatable reading check
Open the primary source rather than stopping at a summary. Read the methods, outcome definitions and limitations, then compare the paper with the claim that led you there. Check publication corrections where available. PubMed provides identifiers and search tools; being indexed is not an endorsement.
- Write one sentence describing what was measured.
- Write a second sentence describing what was not tested.
- Keep an unresolved identity, formulation or route difference visible.
- Do not turn a selected set of papers into a claim that all research has been reviewed.