On this page
A source list can look complete while the evidence behind it is not. Two files may share a title but represent a working paper and a published article. A DOI may resolve while the local PDF is a different version. A page capture may be present while the relevant table is outside the saved text layer.
Source verification is the chain from a citation to the identity, version, held artifact, evidence locator, and permitted use of the work. It is a reproducibility control for sources, not a guarantee that the source is true.
The defensible evidence-review article owns the decision to stop when evidence is not strong enough. The claim-entailment article owns the sentence-to-source support test. This page owns the record that tells another reader which source and version were actually used.
What does source verification mean?
Treat each step as a separate event:
| Step | Question | Evidence of completion |
|---|---|---|
| Discovery | Where did the work or page first appear? | Search or catalog lead with date |
| Retrieval | Could the cited object be obtained lawfully? | URL, DOI, repository path, or capture record |
| Identity | Is this the intended author, title, date, and work? | Bibliographic fields and identifier |
| Version | Is it published, accepted, working, corrected, or captured? | Version statement and publication record |
| Receipt | Do we hold the complete object in that version? | Local file, checksum, and register row |
| Evidence | Which passage, table, or result supports the claim? | Locator and bounded extract |
| Recheck | Could identity, access, or status have changed? | Recheck date and result |
Table 1What does source verification mean?
Source: Table from this essay. Sources and interpretation are given in the article.
The word version does real work. A version of record is a named publication version, not a general synonym for “a PDF.” An accepted manuscript can be useful and still differ from the published text. A page capture can preserve a live page and still need a date and content check.
Does a DOI prove the source claim?
No. A DOI is valuable source identity evidence. It does not prove that the file beside it is the cited version, that the page range is correct, that a result appears in the work, or that the work supports the sentence written by the analyst. Those are separate checks.
Greenberg’s study of one biomedical citation network is a useful warning about why the distinction matters. He documents citation bias, amplification, and invention in that bounded network, including the conversion of a hypothesis into a statement presented as fact through citation alone (Greenberg, 2009). The finding is not a universal error rate. It is a reason to keep the source object and the claim object separate.
What does a verification register look like?
The following rows are synthetic. They show source-control states, not a ranking of sources or a private bibliography.
| ID | Citation and identity | Version and receipt | Text and locator | Claim status | Recheck or disposition |
|---|---|---|---|---|---|
| S-01 | Article, authors, year, DOI agree | Published version held; checksum recorded | Full text; results section located | Verified for bounded result | Recheck 2027-03-01 |
| S-02 | Title matches; author list incomplete | Accepted manuscript held | Methods text located; publication pages differ | Use only as accepted version | Find version of record |
| S-03 | DOI resolves to intended article | Local file is a working paper | Abstract only; result not checked | Identity verified; claim open | Do not cite load-bearing result |
| S-04 | Page title and retrieval date agree | Page capture held; source page changed later | Relevant paragraph preserved | Verified for dated page statement | Recheck if page is reused |
| S-05 | Citation matches a retracted work | Retraction notice held | Original result located | Blocked for new claim | Record retraction and cut claim |
| S-06 | PDF opens but text layer is empty | Artifact retained as page images | Locator requires image check | Incomplete text verification | Inspect page image before use |
Figure 1The synthetic source-verification register
The rows are illustrative. Identity, version, receipt, text, locator, claim status, and recheck are separate states.
Source: Author's synthetic register grounded in the local receipt protocol and Greenberg (2009); all rows and dates are illustrative.
S-03 is the most common trap: identity can be correct while the claim remains open. S-05 shows why a source can be a real publication and still be blocked for new use. S-06 shows why a file’s existence is not evidence that its text has been checked.
Which verification failures need different responses?
- Identity failure: resolve the intended work before extracting claims.
- Version failure: label the held version and do not silently cite a different one.
- Receipt failure: mark the source part-only or none; do not upgrade it from an abstract.
- Locator failure: find the exact passage, table, or page before the claim is load-bearing.
- Status failure: check retraction, correction, and supersession before reuse.
- Text-layer failure: inspect page images or obtain a usable text layer; a failed grep is not a source contradiction.
The response is not always “find another citation.” Sometimes it is to narrow the claim, preserve the source as provenance only, or record that the question is unresolved.
How should a team verify a source before publishing?
- Write the intended citation and the exact version in the source register.
- Resolve identity through DOI, URL, repository, or publication record.
- Preserve the held artifact and record its retrieval date and checksum where available.
- Confirm that the held version matches the version named in the article.
- Extract the relevant passage and record a page, section, table, or dated-page locator.
- Run the claim-entailment test on the sentence that will cite it.
- Check retraction, correction, access, and text integrity before the release gate.
The final status should be explicit: verified for this claim, verified only for identity, incomplete, superseded, blocked, or not held. Source verification makes a citation reproducible. It does not make a proposition true, a model causal, or a literature search complete.
References
- Greenberg, S. A. (2009). How citation distortions create unfounded authority: analysis of a citation network. BMJ, 339, b2680. https://doi.org/10.1136/bmj.b2680