Reproducibility is a publication requirement. Every article ships the code and data needed to rerun it. Where the data cannot be public, as they often cannot be in health research, a verified, privacy-preserving surrogate takes their place, and where a use agreement or a consent forbids even a derivative, a documented restriction with its access route does. "Available on request" is not sufficient.
The reproducibility package
Every submission carries:
- an executable source or code, with the one entry-point command that regenerates the results
- the data, or a privacy-preserving surrogate where the real data cannot be
released
i
A verified synthetic twin of the restricted data, built in a secure enclave: artificial records that reproduce the paper's results without exposing any real one. "Verified" means checked on two sides. It must reproduce the analysis, which is the utility test, and it must pass a disclosure-risk audit, which is the privacy test. Clearing both is what lets a reader rerun the surrogate and trust what comes out. The founders of FAIR Press Journals work in statistical disclosure control. Raw data never leave the enclave and never touch a cloud AI model.
- a pinned environment, meaning a lockfile, requirements file or container, confirmed before acceptance
- determinism information, meaning seeds and any known sources of run-to-run variation
We assess reproducibility by re-execution. Where the article type carries runnable artifacts, the analysis is rerun from the package before publication, and the published article states the outcome as its badge. What is rerun is the analysis, never a re-recruited cohort or a repeated trial, and the badge never implies otherwise. Papers with no runnable artifacts (pure theory, for example) instead state that no code or data apply, and list any machine-checkable proof or symbolic-computation files where those exist.
Everything in the package is submitted de-identified, like the manuscript itself. What that means concretely, and what happens if identifiable data is nonetheless found in a submission, is set out under author guidelines.
An availability statement on every article
Each published article carries a data-and-code availability statement, composed at submission from what the authors actually provide, so the statement and the artifacts cannot drift apart.
When the data cannot be public
Patient records, disease and health registers, linked administrative data, cohort and biobank data held under a use agreement, electronic health records released only inside a safe haven. All of these are a real and legitimate constraint, and none of them is an exemption from verification. Here is the route.
- The restriction is documented at submission, covering what the data are, why they cannot be released, under what regime they were held, and the route by which another researcher could apply for the same access.
- Where feasible, the authors release a safe companion dataset, either
carefully anonymised real records or a synthetic surrogate, chosen to fit the data's actual
disclosure-risk structure. The journal offers this as an opt-in,
separately priced service run by the founders of FAIR Press Journals, whose research field is
statistical disclosure control.The service is in preparation and not yet accepting engagements. Until it opens, authors build their own safe dataset or commission one elsewhere.
i
Submitting never requires it, and the article charge never includes it. The founders of FAIR Press Journals build the safe companion dataset inside a secure enclave and quote it per dataset. The raw sensitive data never leave that environment and never touch a cloud AI model, and choosing the service has no bearing on the accept or reject decision.
- The published result is then re-executed on the safe dataset, so the verification claim rests on an artifact readers can actually download.
- Raw sensitive data are handled under a data processing agreement, are never published, and are never processed by any cloud AI model. They stay in a controlled environment.
That last rule is about the datasets this service handles. It is not a claim about your manuscript, which is assessed by the disclosed review engine on a commercial model API under a data processing agreement with EU standard contractual clauses (see peer review). The two channels are different, and this journal will not blur them.
An article on this route carries the Privacy-limited badge. A declared restriction, documented with its access route, is a state of the evidence and not a fault. That badge holds whatever the surrogate rerun produced, because an article carries one badge and never a stack. Where a surrogate exists and the analysis reran on it, the article's reproducibility panel says exactly that in words.
Where no surrogate can lawfully be built, because some use agreements and consents forbid derivatives including synthetic ones, the route still works. Document the restriction and its access route, ship the analysis code and the derived tables the results rest on, and the article publishes on the same route with the same badge. What is never acceptable is a bare "available on request".
The service and the editors, and the conflict rule
This journal is edited from inside the imprint whose owners run that optional service, so the rule that governs it is printed here rather than left to trust.
- Where an article's authors engaged the service, that engagement is disclosed on the published article.
- The editor who handles a paper takes no part in that paper's service decision, and whoever decides or prices a service engagement takes no part in the editorial decision on the paper it concerns. This journal has one editor today, so if that split cannot be made for a particular paper, because no uninvolved editor is available to handle it, the submission is declined and the authors are told why, instead of being handled by someone with a stake in it.
- Buying the service, or declining it, has no bearing on accept or reject, and the deciding editor does not see payment status (see fees & waivers). Authors are free to build their own safe dataset, or to have a third party build it. What the journal requires is a safe dataset the analysis demonstrably reruns on, with an honest account of how it was made and what it does and does not protect. A purchase from us is never part of that.
Limits
Anonymisation lowers disclosure risk without removing it. Synthetic data are not private by construction, and they face the same disclosure-risk audit as anonymised records. And some analyses, such as rare diagnoses, small-area geography, exact event dates or extreme quantiles, may fail to reproduce faithfully on any safe dataset. When that happens the article says so. Overclaiming privacy is itself a harm.
Licences and preservation
Code and data publish under open licences by default (see open access & licensing), and the package will be preserved alongside the article (see archiving & preservation).