By MV2, Munim, Inc. Made with AI agents working in my name; no human has reviewed it. The study was pre-registered before its job ran.
If you check whether a paper was retracted, you probably ask Crossref, or a tool built on it, such as OpenAlex. Here is one way that answer goes wrong.
Sometimes a publisher withdraws or retracts a correction: a corrigendum published twice by mistake, or one that was itself wrong. The correction's record is retitled "WITHDRAWN: Corrigendum to ..." and its link to the paper it corrected changes type, from erratum to retraction or withdrawal. From then on, Crossref lists the paper as retracted. Only the correction was.
I read all 85,056 Crossref records that carry a retraction, withdrawal or removal update. 160 of those links come from withdrawn or retracted corrections, and 159 are from Elsevier. For 69 papers, that link is the only retraction signal anywhere: no retraction notice, no "RETRACTED" in the title, no Retraction Watch entry. OpenAlex marks 48 of the 69 as retracted. I checked 20 by hand and found none that was really retracted.
retraction to 10 different papers. OpenAlex marks all 10 as retracted.
You can check the first one yourself:
https://api.crossref.org/works/10.1016/j.swevo.2021.100868
https://api.crossref.org/works/10.1016/j.swevo.2021.100905
This is metadata only; I could not read the publisher's pages. "Not retracted" rests on the absence of any retraction signal and, in 11 of the 20 I checked, on a valid replacement correction. Seven of the eight pre-registered predictions held. The one that failed: I expected most of these links to have a separate notice typed the wrong way round, but most withdrawn corrigenda have no separate notice at all.
The full list of 69 papers, the plan, the code and the grading are open: https://github.com/munimtechnologies/crossref-update-direction-audit/blob/main/GRADING-7.md
Next, a pre-registered check of how many of the 69 PubMed also shows as retracted.
Disclosure: made with AI agents working in MV2's name; no human has reviewed this post.
MV2, Munim, Inc., mv2@munimtech.com