The seven verdicts: how each is reached and what to do next
What Benthic compares before it calls a reference verified, deviating or invented, why books, chapters and very new papers need care, and how to repair each case.
A check comes back with “Deviates” next to a paper you know well, and “Uncertain” next to a chapter you photocopied in the library last week. Neither verdict means Benthic doubts you. Each one says what the comparison found, and once you know what is compared, you know what to do next.
What goes into a verdict
Benthic compares the reference as cited with the work it finds in the registries (how it finds that work is the subject of checking a reading list). Five things count:
- Title. It counts most. Case, accents and punctuation are ignored, and a title that appears whole inside the other one (a missing subtitle, say) counts as the same.
- Authors. One shared surname is enough, because cited author lists are often cut short with “et al.”.
- Year. One year either way still matches, because a preprint and its printed version are often a year apart. For arXiv papers the year of the first version counts.
- Identifier. Whether the DOI exists, and whether it belongs to the work with that title.
- Retraction metadata. The retraction flag in OpenAlex and the retraction notices in Crossref. A retracted work gets the verdict Retracted, however well everything else matches.
Real, citable: Verified and DOI corrected
Verified means the identifier led to a work, or a title search found one clearly, and title, authors and year agree. There is nothing to repair.
DOI corrected has two routes. Either the DOI does not exist at all, or it exists but belongs to a work with a very different title. In both cases Benthic searches by title, and if it clearly finds the cited work under another DOI, it puts that one in. The example from the product page is LeCun, Bengio and Hinton (2015), “Deep learning”, cited with 10.1038/nature14538. That DOI belongs to an item called “Addendum”; the paper itself is 10.1038/nature14539. Take the corrected DOI into your list. Benthic shows it in the report and fetches the paper with it.
Look it up: Deviates and Uncertain
Deviates means the work exists, but its title, authors or year differ from the reference. The note says which, for example “Work found, but the year differs from the reference.” Typical causes are a year that is off by more than one (an online-first date, a later edition), a misspelt name or a title quoted from memory. Take the details from the work that was found, once you have checked that it is the edition or version you read.
Uncertain covers the cases where Benthic sees something but cannot settle it:
- A title search finds only a similar title: “Closest match: … Please check.”
- The DOI is registered, but the registry holds no details to compare.
- A book chapter (“In: … (Eds.)”) that no registry lists: “Book chapter: often not listed in Crossref or OpenAlex. Please check it in a library catalogue.”
- A consulting or business source, such as a Harvard Business Review article, which you check with its publisher.
- A web page or report whose link works, so the content is yours to check.
- Crossref did not answer, so Benthic will not call the source invented: “Not found, but Crossref did not answer. Please check again later.”
The repair is a look at the original: the library catalogue for books and chapters, the publisher's page for the rest. If the closest match is the work you meant, correct the reference to fit it.
Do not cite: Mismatch, Not found, Retracted
Mismatch: the DOI exists but belongs to a different work, and a title search does not find the cited one. This is what a reference stitched together from parts looks like. The site's invented example, Smith, J. (2019), “Deep learning for protein folding at scale”, Science, carries a DOI that belongs to “Cell Proliferation Without Neurogenesis in Adult Primate Neocortex”. Remove it, and if the claim it was meant to support matters, find a real source for it yourself.
Not found: neither the identifier nor a title search in Crossref, OpenAlex, arXiv, Semantic Scholar or Open Library leads to a matching work. Usually the reference is invented. If it carries a web link that does not work, the verdict is Not found as well. It is a strong signal, not proof: search your library catalogue once before you delete the entry.
Retracted: the work exists and was withdrawn. Wakefield et al. (1998) in The Lancet is the well-known case. It belongs in no reference list, unless the retraction itself is your subject.
Books, chapters and very new papers
The registries are strongest for journal articles with a DOI. Elsewhere, the verdicts lean towards caution:
- Books rarely have a DOI. Benthic searches Open Library by title and first author, and a match reads “Found by title (Open Library).”
- Chapters in edited volumes are often missing from Crossref and OpenAlex. Semantic Scholar lists some; when nothing turns up, the verdict is Uncertain rather than Not found.
- Very new papers take a while to appear in the registries, and a preprint from last week can come back Not found. Check it again later.
- Preprints that have since been published often carry a different year from the journal version. The one-year allowance covers the usual gap.
Verdicts are well-founded pointers, not expert opinions, as the PDF report notes at its end. With Deviates and Uncertain, the look at the original is part of the job. The short version of all seven is on the product page.