How we check
Last updated: 8 September 2026
Graze runs the same checks on every paper and shows you which ones fired. None of them tell you whether a paper is any good — that part is still yours.
The three results
Field-Tested means the publisher records the work as peer-reviewed, the retraction check ran and found nothing, at least one independent source corroborates either the journal or the authors, and nothing we check raised a concern.
Unverified means some check couldn't be completed — usually a missing DOI, or authors we couldn't resolve to any registry. It is not a mark against the work. Preprints land here too, because they haven't been through peer review yet, which is a fact about their stage rather than their quality. Unverified results are never hidden and never pushed down the page.
Flagged means something specific and serious: a retraction on record, a match against a predatory-publisher list, or a journal reported as hijacked. Flagged results are held back from the default list, and you can always choose to see them.
Not every check counts the same
This is the part worth knowing, because it's where most credibility scores go wrong. Our checks are split into three weights, and the split is deliberate.
- Decides on its own. A retraction, a predatory-list match, a hijacked journal, or sources disagreeing about the paper's basic details. Any one of these flags a paper regardless of everything else — a retracted paper with ten thousand citations is still retracted.
- Moves it between verified and unverified. Index listings, the peer-review type, fee and review-process disclosure, author identities. None of these can flag a paper on their own.
- Context only. Citation counts. Shown so you can judge for yourself, and deliberately unable to change the result — otherwise recent or specialised work would look disreputable for being new or niche, which is exactly the research Graze exists to surface.
The checks themselves
Retraction — Crossref
Checked against the Retraction Watch records Crossref distributes, per DOI. This covers retractions those records know about. It is not a guarantee that no concern exists about a paper — only that no retraction notice has been filed.
Predatory-publisher list
Matched against the Stop Predatory Journals list, synced weekly. The list identifies journals by name, and names are not identifiers — hundreds of its entries are single generic words that are also the names of entirely legitimate journals. So a match is only acted on when the name is specific enough to identify one publication and no reputable index disagrees. Everything else is held for a person to look at, and shown to you with a note rather than hidden.
Peer review — Crossref and OpenAlex
Taken from the publisher's own declared work type. It records what the publisher says happened, not an inspection of the review itself.
Journal indexes — DOAJ, MEDLINE
DOAJ vets peer-review practice, licensing and editorial governance before listing a journal, and covers open-access journals only. MEDLINE is selected by the US National Library of Medicine and covers biomedicine.
Absence from either means very little. Nature isn't in DOAJ, because it isn't open access. A good law review isn't in MEDLINE, because it isn't medicine. We never treat "not found" as evidence of anything.
Author identity — ORCID and ROR
An ORCID iD is a persistent identifier a researcher registers for themselves. A ROR id means the author's stated institution resolves to a real registered organisation rather than an unmatched string.
Both are limited, and the limits matter: having an ORCID doesn't vouch for the work, we don't check whether that ORCID record has any publications on it, and a resolved institution confirms the institution exists — not that the person works there. Plenty of real researchers, especially before 2012, have no ORCID at all, so we never count its absence against anyone.
Fees, review process and turnaround — DOAJ
A journal's own published figures. Disclosing an article fee up front is a good sign; a stated turnaround of two weeks or less is called out, because peer review that fast is a common pattern among paper mills.
What none of this tells you
Every check above is about a paper's status — where it was published, whether it was withdrawn, whether the people and places are real. None of it says whether the methods are sound, the statistics hold, or the conclusions follow.
A peer-reviewed paper in a well-indexed journal can still be wrong, and an unverified preprint from a journal nobody has indexed can be excellent. That judgement is the reading, and no badge replaces it. It's why the Chew on this summary always ends by sending you to the source.
See also the Terms & Conditions, which says the same thing in the language a lawyer would want.