Reproducible product case
The DOI is real. The paper title is not.
This controlled case uses a real Nature DOI attached to a deliberately fabricated title. It documents the exact input used in the product test suite, the record returned by scholarly sources and why the checker reports a metadata mismatch instead of a verified reference.
Free reference checker for academic citations.
Reviewed by SolveUnit Scholar Engineering and Research on .
What makes this case reproducible
- The submitted citation is printed below exactly as tested.
- The matched DOI, paper metadata and independent source links are public.
- The expected status and confidence are asserted by the automated test suite.
From convincing citation to visible mismatch
1. Start with a deliberately wrong citation
The test input names John Jumper, the journal Nature and the real DOI 10.1038/s41586-021-03819-2, but replaces the paper title with “A completely fabricated study about life on Mars.” The mixture is plausible-looking and intentionally unsafe to cite.
2. Resolve the identifier before searching by keywords
The DOI resolves to “Highly accurate protein structure prediction with AlphaFold,” published in Nature in 2021. Identifier resolution gives the checker a specific record to compare instead of a merely similar search result.
3. Compare the submitted fields separately
The first author, year, venue and DOI agree with the record. The title does not. Agreement in several fields must not hide a direct conflict in the field that identifies what the paper is about.
4. Let the conflict override the score
The implemented rule returns metadata_mismatch with confidence 55. It does not label the citation verified simply because the DOI is valid or because four source fixtures support the underlying record.
5. Return the matched record without silently repairing the input
The result preserves the submitted citation, exposes the title difference and provides a formatted citation for the actual AlphaFold paper. The reader can see both what was supplied and what the DOI really identifies.
6. Keep database availability separate from identity
In this fixture Crossref, Semantic Scholar, PubMed and Europe PMC support the same record while OpenAlex is unavailable. An unavailable source does not erase the evidence returned by the other sources, and it is not counted as a negative match.
7. Finish with a human source check
The mismatch proves that this citation text does not accurately describe the DOI record. It does not prove that the AlphaFold study is correct or that it supports a particular claim. Those questions require reading the paper.
Five scholarly databases
Crossref
Fixture checked: DOI, title, first author, year and journal agree with the AlphaFold record.
Semantic Scholar
Fixture checked: an independent paper record supports the same identity.
PubMed
Fixture checked: PMID 34265844 supports the same biomedical publication.
Europe PMC
Fixture checked: the life-science index supports the same PMID and paper identity.
OpenAlex
Fixture unavailable: absence from this check is recorded as unavailable, not as a disagreement.
A fabricated title attached to the AlphaFold DOI
Submitted: Jumper, J. et al. (2021). A completely fabricated study about life on Mars. Nature. https://doi.org/10.1038/s41586-021-03819-2
Result: Metadata mismatch · confidence 55. The DOI is valid, but the submitted title conflicts with the record.
Matched record: Jumper, J. et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583-589. https://doi.org/10.1038/s41586-021-03819-2
Cases that protect the status logic
These deterministic cases are covered by the reference-verification test suite. No unmeasured accuracy rate is claimed.
Metadata mismatch · confidence 55
Input: Fabricated title plus the real AlphaFold DOI.
The title conflict overrides the valid DOI and agreeing author, year and venue.
Verified · confidence 100
Input: Correct AlphaFold title, year, venue and DOI.
The important supplied fields agree with the record and the exact DOI identifies it.
Metadata mismatch
Input: Correct title and DOI with the year changed to 2020.
The year difference remains visible even though the title and identifier agree.
Incomplete
Input: No result while only OpenAlex answered.
A source outage is not converted into evidence that a citation is fabricated.
What this check cannot prove
- This is a deterministic behavior case, not an accuracy benchmark or coverage study.
- Live provider availability and metadata can change after the documented review date.
- A matching publication record confirms identity, not methodological quality or support for a surrounding claim.
Frequently asked questions
Why is the confidence 55 if the DOI is exact?
The score is metadata agreement, not DOI validity. A direct title conflict applies a penalty and forces metadata mismatch.
Why not automatically replace the fabricated citation?
Silent replacement would hide the error and could cause a reader to cite a paper they never intended to use. The submitted and matched values remain separate.
Can I reproduce the check?
Yes. Paste the exact controlled input into the free reference checker, then open the DOI and PubMed records linked on this page.
Does this case show that all fabricated references can be detected?
No. It demonstrates one protected behavior: a valid identifier cannot override conflicting title metadata.
Primary sources
- Nature: Highly accurate protein structure prediction with AlphaFold
- DOI record: 10.1038/s41586-021-03819-2
- PubMed record: PMID 34265844
- Crossref REST API documentation
- SolveUnit Scholar verification methodology
Why publish a controlled case
A checker should show how it behaves when an identifier is real but the surrounding citation is false. The case is intentionally synthetic, clearly labeled and linked to real source records; it is not presented as a user submission.