For most of the history of scholarly publishing, a research output has been a fixed thing. A paper is written, reviewed, published and then, save for the occasional correction, frozen: the version that appears is the version that endures, cited and read in exactly the form it was issued. There is a real virtue in this fixity — a stable, citable record everyone can refer to with confidence. But there is also a tension, because knowledge does not stand still. New evidence arrives, errors are found, methods improve, and a static document published years ago may no longer reflect what is known. A growing class of living and versioned research outputs tries to resolve this tension by allowing outputs to be updated over time while remaining citable and trustworthy. This article explores those outputs and the challenges they raise, drawing on the research outputs domain of the CASRAI Dictionary.
Living systematic reviews
The clearest example of a deliberately dynamic output is the living systematic review. A systematic review gathers and synthesises all the available evidence on a question, and it is enormously valuable — but it has a built-in problem: the moment it is published, it begins to go out of date, because new studies keep appearing. In a fast-moving field, a review can be obsolete within a year or two, yet it may continue to be cited as authoritative. A living systematic review addresses this by being continuously or regularly updated as new evidence emerges, rather than being conducted once and left to age. The review becomes an ongoing process — a maintained synthesis that keeps pace with the literature — rather than a one-off snapshot. This is invaluable in areas where keeping current matters most, but it changes the nature of the output: it is no longer a fixed document but a living one, and that has consequences for how it is cited and trusted.
Versioned preprints
A different but related development is the versioned preprint. Preprints — research papers shared publicly before, or alongside, formal peer review — are inherently dynamic: an author posts an early version, receives feedback, revises, and posts a new version, often several times. Preprint servers handle this through explicit versioning, so that version 1, version 2 and so on each exist as distinct, citable entities, and a reader can see both the latest version and the history of how the work evolved. This is honest and useful: it shows the work developing, and it lets a reader cite the specific version they actually read. But it also means that “the preprint” is not a single thing — it is a series of versions, and which one a citation refers to genuinely matters.
The version-of-record problem
All of this raises a fundamental question that the traditional model never had to face squarely: in a world of multiple versions, what is the version of record? The version of record is, classically, the definitive, citable version of a work — the one the scholarly record points to as authoritative. When outputs were fixed, this was simple. When an output exists in many versions, or is continuously updated, several questions become pressing:
- Which version is authoritative — the latest, or the one a given reader relied upon?
- How do citations stay precise when the thing being cited keeps changing?
- How is the history preserved so that a claim made on the basis of an earlier version can still be checked against that version?
- How does a reader know they are looking at the current state of a living output, rather than a superseded one?
These are not merely technical questions; they go to the heart of what makes the scholarly record reliable. A record that changes without trace would be untrustworthy; a record that cannot be updated would be inaccurate. The challenge is to allow updating while preserving citability and history.
How DOIs handle versions
The infrastructure that makes versioned outputs workable is the persistent identifier, and in particular the way DOIs can be assigned to versions. A common and powerful pattern is to mint a DOI for each specific version of an output and a separate “concept” or top-level DOI that always points to the latest version of the work as a whole. This gives the best of both worlds: someone who wants to cite exactly what they read can cite the version-specific DOI, confident it will always resolve to that exact version; while someone who wants to point readers to the current state can use the concept DOI, which follows the work as it evolves. Versioning at the identifier level is what lets a living output be both stable (each version is fixed and permanently citable) and dynamic (the work as a whole keeps moving forward). It reconciles the apparent contradiction between fixity and change.
Continuously updated outputs more broadly
Living reviews and versioned preprints are the prominent cases, but the underlying pattern — outputs that are maintained and updated rather than issued once and frozen — appears elsewhere too: datasets that grow and are re-released, software that moves through versions, guidelines and protocols that are revised as practice changes. In each case the same principles apply: clear versioning, persistent identifiers for both specific versions and the evolving whole, and transparent records of what changed and when. The broad taxonomy of modern research outputs increasingly has to accommodate things that change over time, not just things that are finished and fixed.
A shared vocabulary for versions and outputs
For versioned and living outputs to work across repositories, publishers and citing systems, the concepts involved must be described consistently — what a version is, how it relates to the work as a whole, which is the version of record, and how the relationships between versions are expressed. Inconsistency here breaks exactly the citability that versioning is meant to preserve. That consistency is what the CASRAI Dictionary supports: a shared vocabulary so that version and output information is understood identically wherever it appears. And because maintaining a living output over time is genuine, ongoing contribution, the work can be described in the same framework used for every other — the CRediT taxonomy and its full set of contribution roles. The scholarly record is learning to do something it never used to: stand still enough to be trusted while moving enough to stay true.