Research integrity is usually discussed as a matter of editorial judgement: the editor who spots a manipulated image, the reviewer who questions a statistical method, the integrity officer who investigates a complaint. Those people remain indispensable. But much of what determines whether a reader can trust a published article is decided long before any of them open a file, and long after the article goes live. It sits in the publishing infrastructure: the metadata a journal registers, the identifiers it collects, the records its systems keep and the way it signals that something has changed.
A well-designed publishing system cannot make research correct. What it can do is make the history of a scholarly work visible and checkable: who did what, which data underpin the findings, how the manuscript was handled, and whether the article has since been corrected, flagged or retracted. When that information is missing, or exists only in an editor's inbox, every downstream user — readers, indexers, librarians, funders, other authors — has to take the article on faith.
This explainer sets out the main building blocks of integrity by design, the community standards that already exist for each, and a practical checklist for publishers. It is deliberately about infrastructure rather than detection; JCF Media has covered paper mills and the rise in retractions separately.
The version of record, and what happens after it
The version of record is the publisher's final, authoritative form of an article. The problem is that it rarely stays final. Errata, corrections, expressions of concern and retractions are routine parts of scholarly publishing, yet for years the most common failure was not the decision itself but its communication. A retraction notice might appear on the journal site while copies of the original PDF, aggregator records and reference-manager entries carried on as if nothing had happened.
Crossref's Crossmark service was designed to address this. Publishers register update metadata linking a correction or retraction notice to the original DOI, and readers who click the Crossmark button — on the HTML page or inside a downloaded PDF — can see the article's current status. Crossref now lets any member deposit update metadata and register an update policy whether or not they display the button, which matters because machines read the metadata even when humans never see the badge. Crossmark carries no additional fees.
The coverage of that metadata has also improved. In September 2023 Crossref acquired the Retraction Watch database, which records tens of thousands of retractions, and made it openly available. Retraction Watch continues to maintain the data, while publishers still register their own retraction notices with Crossref. The practical effect is that tools checking a reference list against known retractions now have a shared, open source to draw on.
In June 2024 NISO published a Recommended Practice on the Communication of Retractions, Removals, and Expressions of Concern (CREC, NISO RP-45-2024). It sets out how metadata about the original article and the accompanying notice should be created, transferred and displayed, so that an article's status is clear both to human readers and to automated systems. For publishers, CREC turns a general aspiration into a specification that platform vendors and production teams can implement.
Knowing who the authors are, and what they did
Identity is the second layer. ORCID gives researchers a persistent identifier that follows them across institutions and name changes; ROR does the same for research organisations, replacing free-text affiliation strings that are notoriously inconsistent. When a submission system collects authenticated ORCID iDs and ROR identifiers, and passes them into published metadata, editors gain a far better basis for checking that authors are who they say they are and that affiliations are real.
Identifiers do not prove that a person did the work, and they are not a defence against fabricated identities on their own. They do, however, make patterns visible, which is precisely what integrity teams need when investigating coordinated misconduct.
Contribution is a related question. The Contributor Roles Taxonomy, CRediT, became a formal standard in 2022 as ANSI/NISO Z39.104-2022. It defines 14 roles, from conceptualisation and data curation to formal analysis, software and writing. Recording roles in structured form, rather than in a narrative paragraph, makes authorship claims easier to compare with the actual content of a paper and easier to query at scale.
Connecting articles to their data
A data availability statement is only as useful as the link behind it. DataCite, the DOI registration agency for datasets and other research outputs, lets repositories record typed relationships between a dataset and an article — for example, that a dataset is supplement to a particular paper, or that one version is new version of another. When publishers deposit the corresponding links in Crossref, the connection becomes discoverable from both ends.
For integrity purposes, linked and persistent data matter more than polished statements. A reader trying to check a figure needs to reach the underlying files, and an editor handling a complaint needs to know which version of a dataset the article relied on.
Screening, guidance and the editorial audit trail
Infrastructure also includes the checks run during peer review. The STM Integrity Hub, developed by STM Solutions, is a shared cloud environment in which participating publishers can screen submitted manuscripts for signs of paper-mill activity and other integrity concerns. Its value lies partly in pooling signals across publishers, since some misconduct — duplicate submission, for instance — is invisible to any single journal.
Guidance sits alongside tooling. COPE's retraction guidelines and its wider body of flowcharts and case discussions remain the reference point for how editors should handle concerns, communicate decisions and treat authors fairly. Infrastructure should make following that guidance easier, not replace it.
The least glamorous element may be the most important: an audit trail of editorial decisions. A submission system that records who invited which reviewers, which reports were received, what the editor decided and when, gives a journal something to examine when a published paper is questioned. Without that record, investigations depend on memory and personal email. With it, a publisher can establish whether a decision followed its own process — and, where it did not, show what went wrong. Concerns about AI in peer review make this documentation more, not less, important.
A practical checklist for publishers
- Register Crossref update metadata for every correction, expression of concern and retraction, and link each notice to the original DOI.
- Display current status on the article page and in PDFs, using Crossmark or an equivalent, and keep the original article accessible but clearly marked when retracted.
- Review platform display and metadata feeds against the NISO CREC Recommended Practice.
- Check reference lists at submission against the openly available retraction data now distributed by Crossref.
- Collect authenticated ORCID iDs for authors and, where possible, reviewers; record affiliations with ROR identifiers.
- Capture contributor roles using CRediT and include them in published metadata.
- Require data availability statements that point to persistent identifiers, and deposit article–dataset relationships.
- Participate in shared screening infrastructure such as the STM Integrity Hub where appropriate, and document what each check does and does not detect.
- Align editorial policies and complaint handling with COPE guidance, and publish those policies.
- Keep a complete, retained audit trail of editorial decisions in the submission system.
Conclusion
None of these measures is new or exotic. Most are established standards maintained by community organisations, and many cost little beyond implementation effort. What is often missing is the decision to treat them as core design requirements rather than optional extras added after a crisis. Publishers that build integrity into their systems will still face misconduct, errors and difficult judgements. The difference is that when those problems arise, the record will show what happened — and readers will be able to see it too.
