A lot has landed since the July update. The headline items: you can now format references in essentially any citation style, results get a scoresheet so you can see the state of a whole reference list at a glance, and we spent most of the summer hand-checking RefRunner’s output against hundreds of published papers and fixing everything we got wrong.
Add more easy-to-access citation styles—plus the entire Zotero catalog
The style picker no longer has just the most common styles. We added a bunch of publisher styles (38 styles across 8 major publishers — Elsevier, Taylor & Francis, SAGE, etc).
But wait! That’s not all! If those are not what you need, there is a search-as-you-type style finder that covers the entire Zotero style repository — over 10,000 journal and publisher styles. Type your target journal’s name and pick it; RefRunner reformats your list in that journal’s style. It’s now conceivable that someone other than me can add a style from Zotero’s repository.
An interesting finding is that Wiley is now publishing all journals (that we checked) in their own in-house Chicago-like style rather than the APA, Vancouver, or whatever style is expected by researchers reading those journals. I’ve heard a couple authors I know complain about this, but haven’t seen any public outcry.
The Score Sheet
Results now open with a row of tallies: ✓ verified, 
needs review, 
DOIs verified, 
missing DOIs found, 
web refs, and not indexed (conference talks, dissertations, and other grey literature that bibliographic databases don’t cover — for those, “not found” is expected, not suspicious). Each tally is a filter — click “needs review” and you see only the references that need a human look.
Two related honesty fixes: formatting-only differences no longer count as “needs review” (a comma is not a scandal), and the DOI tally now counts every DOI that checked out, so the numbers add up the way you’d expect.
A summer-long accuracy push
We downloaded corpora of published papers from eight publishers — PLOS, Elsevier, Emerald, Wiley, SAGE, Taylor & Francis, and others — ran every paper through RefRunner, and hand-checked the flags. This turned up a bunch of parsing bugs, which are now fixed. It’s still possible that a reference we can’t find exists, but that number is substantially reduced. It’s very unlikely that a reference we do not flag is fake, or has the wrong DOI.
We don’t expect that many people will need to process PDF files—it’s always better to use a .docx or .odt file, but this was the only way to get a big corpus of test papers. Better PDF parsing enabled us to find and fix a bunch of issues:
- Numbered reference lists (Vancouver, IEEE, AMA) parse much better: run-together entries are split apart, and numeric citations like [12] are matched to reference 12 by its list number.
- Corporate authors — WHO, UNESCO, OECD and friends — are recognized in both reference lists and in-text citations instead of being chopped into fake surnames.
- Names survive PDFs better: accents that drifted off their letters, multi-word surnames, “Jr.” suffixes, and hyphenated names split across lines are all handled.
- ALL-CAPS journal names and abbreviated journal titles now match their records.
- “n.d.” and “in press” citations match their reference entries.
- Books and chapters: Vancouver-style book imprints parse, editors resolve via ISBN, and useless library-catalog URLs no longer show up as book links.
- A year that appears inside a DOI string is no longer mistaken for the publication year.
Fewer false "not found"s
When Crossref doesn’t know a DOI, RefRunner now falls back to asking doi.org directly, so valid DOIs from smaller registries resolve instead of leaving it up to you to check. Rate-limited or flaky responses from Crossref and Google Books are retried automatically. And a reference that parsed badly can no longer throw away a DOI that had already resolved correctly.
The point of all this: real problems stand out
With our own noise cut down, genuinely bogus references — the kind AI writing assistants fabricate — stand out clearly. In one audit of a published paper, RefRunner surfaced what turned out to be 22 hand-verified faulty references; comparable papers from the same journal scored essentially clean. That contrast is exactly what the tool is for: when RefRunner is quiet, the list is in good shape; when it lights up, something is genuinely wrong.
Smaller things
- Large documents autosave reliably now, saves happen mid-run instead of only at the end, and if a save fails you’ll see it (and it retries) instead of silently losing work.
- SSRN preprints get a “metadata may be stale” badge, and preprint-vs-published-version differences are framed as what they are.
- Datasets and software citations are checked appropriately for their type instead of being held to journal-article rules.
- Non-English papers: we considered adding support for Spanish-language papers, but the parser has a bunch of language-specific regular expressions—even adding support for just using “y” instead of “and” or “&” was going to be a huge job. We’ll have to delay support for other languages for another time.
As always: if RefRunner gets something wrong, hit the “RefRunner got this wrong?” button on the card and we’ll take a look. Several of this summer’s fixes started exactly that way.