Evidence standards
What our labels mean, what can support a claim and what cannot, and the limitations of this dataset that we would rather tell you about than hide.
- Established
- Specialists would state this without hedging. Multiple independent lines of evidence agree.
- Well supported
- Good evidence backs this, though some details remain open.
- Emerging evidence
- Real findings exist, but too few or too recent to be settled.
- Preliminary
- Early results only. Treat as a lead rather than a conclusion.
- Contested
- Researchers actively disagree, and the disagreement is substantive.
- Overturned
- Once accepted, now superseded by better evidence.
- Not enough evidence
- Nobody has done the work needed to answer this properly.
- Popular claim, unsupported
- Widely repeated, with no good evidence behind it.
What can support a claim
Not every source we use is allowed to be the basis of a claim. The distinction is enforced by our validation, not by editorial habit.
Can support a claim
- Nature Portfolio — Peer-reviewed primary research and review articles across Nature, Nature Communications, Nature Ecology & Evolution and Nature Plants.
- American Association for the Advancement of Science — Peer-reviewed primary research in Science and the open-access Science Advances.
- Proceedings of the National Academy of Sciences — Peer-reviewed primary research, with a strong record in animal cognition and neuroscience.
- Cell Press — Peer-reviewed primary research across Current Biology, Cell and iScience.
- Elsevier journals — Peer-reviewed primary research including Animal Behaviour, Applied Animal Behaviour Science and the Trends series.
- Springer journals — Peer-reviewed primary research including Animal Cognition and Behavioral Ecology and Sociobiology.
- The Company of Biologists — Peer-reviewed research in the Journal of Experimental Biology, a core source for sensory biology.
- Public Library of Science — Fully open-access peer-reviewed research; one of the few sources whose text NatureHQ could reuse if it ever needed to.
- PeerJ — Open-access peer-reviewed research with published peer-review history, which makes the review process itself inspectable.
- eLife Sciences Publications — Open-access peer-reviewed research that publishes editorial assessments and reviews alongside the paper, so the strength of a result can be read rather than assumed. A useful home for replications and negative results.
- Wiley journals — Peer-reviewed research including Ecology Letters and New Phytologist.
- The Royal Society — Peer-reviewed research in Proceedings B and Biology Letters.
- Government science agencies — Research documents and advisory reports from national agencies — fisheries, wildlife, geological surveys. Often the only published home for decades-long population monitoring, and usually absent from Crossref, so records sourced here are expected to stay without a DOI rather than to be pending one.
- De Gruyter journals — Peer-reviewed research in long-established specialist journals, notably Mammalia for mammalogy. Narrow use: NatureHQ holds this source for findings that have no home in the larger publishers already covered.
- Brill journals — Peer-reviewed research in long-established specialist journals, notably Behaviour for ethology. Narrow use, as with De Gruyter: held for findings that have no home in the larger publishers already covered.
- University press journals — Peer-reviewed journals published by university presses — The Biological Bulletin, The American Naturalist, Behavioral Ecology and others. Distinct from source:university-press-offices, which covers institutional press releases and is not a source of truth.
- Annual Reviews — Commissioned review journals. Useful where NatureHQ needs the settled state of a field rather than a single result — and treated as synthesis, never as the primary evidence for a specific finding.
- JAMA Network — Medical literature; relevant to NatureHQ where comparative biology meets human medicine.
- Academic book publishers — Scholarly monographs that consolidate a research programme, such as von Frisch on the honey-bee dance language. Cited for what they established; superseded on any point where later research disagrees.
- Public-domain scientific works — Historical monographs out of copyright, such as Darwin and von Frisch. Used for the history of a subject, never as current evidence.
- bioRxiv — Preprints. Legitimate evidence, but never presentable as established: anything from here carries preprint status through to the reader.
- Global Biodiversity Information Facility — Species occurrence records. Authoritative for distribution, with the caveat that occurrence data reflect where people looked as much as where animals are.
- Catalogue of Life — Consensus taxonomic backbone. The intended authority for scientific names and classification.
- IUCN Red List of Threatened Species — The authority for conservation status. Assessments are dated and taxon-specific, and NatureHQ records both alongside the category.
- National and international public health agencies — The authority for what a member of the public should actually do about a health risk from an animal — rabies exposure above all. NatureHQ summarises the shape of that guidance and its jurisdictional variation, and never restates a dose, a schedule or a treatment decision. Those belong to a clinician with the patient in front of them.
Useful, but never the basis of a claim
- Crossref — DOI registration and bibliographic metadata. The intended resolver for verifying NatureHQ study identifiers; it establishes that a paper exists and what it is called, not what it found.
- OpenAlex — Open bibliographic graph covering works, authors and citations. Intended for discovering related and superseding research during enrichment.
- Europe PMC — Literature search across life-science journals and preprints, including retraction and correction notices — the mechanism by which NatureHQ would learn a cited paper had been withdrawn.
- University and institutional press offices — Discovery signal: press releases tell us a study exists and is getting attention. They routinely overstate it, so the underlying paper is always located before anything is written.
- Science and news media — Demand and trend evidence only: what people are reading, and therefore what they are about to search for. Never the scientific basis of a claim.
- A preprint is always labelled as one, and can never be the sole basis of a claim we call established or well supported. Our validation rejects that combination outright.
- A press release is a signal that a study exists, never the evidence itself. We find the paper.
- A newspaper article is evidence about what people are interested in, not about the natural world.
- A claim marked contested must describe the disagreement. A claim marked established or well supported must cite at least one study. These are schema requirements, so a record that breaks them cannot be published.
- We explain research in our own words and cite it. We do not reproduce published text.
NatureHQ’s founding dataset was compiled editorially rather than by querying a bibliographic register. We have since resolved it: every study record was checked against Crossref and OpenAlex, and a DOI is attached only where an exact title, a matching author and an uncontradicted year all agree. Where they did not, the record says so on the page rather than carrying an identifier nobody confirmed.
That check found real errors in our own records — a paper dated two years late, a journal named wrongly, two people credited with work they did not write, and titles we had paraphrased rather than quoted. All are corrected, and each study page now states which register confirmed it and when. Where a reference is legitimately awkward — a book whose only identifier belongs to a later edition, or a venue no register indexes — the page explains the discrepancy instead of hiding it.
We also hold no search-volume data. The priority model behind what we build next records honestly whether demand evidence is measured, directional, inferred or simply unknown — and our schema physically prevents a guess being stored as a number.