A visible standard for every claim.
Our evidence labels summarize how much confidence a reader should place in the central finding—not how exciting, prestigious, or popular a study is.
What we examine
- Study design. Randomization, controls, blinding, prospective plans, and direct measurement generally support stronger inference.
- Evidence base. We consider sample size, population, duration, missing data, and whether the evidence comes from humans, animals, cells, simulations, or observations.
- Effect and uncertainty. Statistical significance alone is not enough. We look for effect size, uncertainty intervals, and practical importance.
- Replication and context. A finding that agrees with independent work deserves more confidence than an isolated result.
- Bias and disclosure. Funding, conflicts of interest, selective outcomes, attrition, and researcher degrees of freedom can change how a result should be interpreted.
What the labels mean
Direct, rigorous measurement or a strong synthesis with limitations unlikely to overturn the central claim. Strong still does not mean absolute certainty.
A credible result with meaningful limitations that could change its size, reach, causal interpretation, or generalizability.
An early human, observational, or small-sample result that raises confidence above speculation but still needs confirmation.
Evidence from animals, cells, organoids, laboratory systems, or mechanisms that has not established human benefit.
A conditional estimate or explanation derived from a model. Value depends on assumptions, inputs, validation, and scenario.
Every label needs a reason
Each research page includes a short study-specific explanation for its classification. Prestige and sample size never determine the label alone. A large observational study may still be moderate, while a small laboratory experiment can provide strong evidence for a narrow physical measurement.
The label applies to the claim
One paper can contain claims of different strength. A laboratory may directly identify a molecule with high confidence while offering a more tentative explanation of how it formed. We classify the central public-facing claim and describe important differences inside the article.
Human, animal, cell and model evidence stay separate
The research subject appears near the top of every page and can be filtered in the database. Animal, cell, organoid and model findings are never presented as established human outcomes.
Corrections and change
Evidence grades can change as studies are corrected, replicated, contradicted, or placed in a stronger review. When a meaningful update changes our explanation, we identify it rather than quietly rewriting history.