Before a manuscript reaches a reviewer, most journals run it through desk screening: a check on scope and completeness, then integrity checks on figures, references, and statistics, then a decision about whether the paper is worth a reviewer's time. Screening protects peer review, which is the scarcest resource in publishing.
What does desk screening actually cover?
Desk screening is the stage between submission and the first reviewer invitation, and it exists because reviewer attention is finite. An editorial office is deciding one question: is this manuscript in good enough shape, and close enough to the journal's scope, to justify asking two or three experts to spend hours on it.
The checks fall into layers that most journals run in roughly this order:
| Layer | What it asks | Typical outcome if it fails |
|---|---|---|
| Scope and format | Is this the right journal, and is the submission complete? | Returned to author, or desk rejected |
| Declarations | Are ethics, funding, conflicts, data availability, and AI-use statements present? | Query to author before proceeding |
| Integrity screening | Do the figures, references, and statistics hold up to automated checks? | Query, or escalation to an integrity specialist |
| Editorial judgment | Is the advance sufficient for this venue? | Desk rejection without review |
The integrity layer is the one that has changed most in the past few years. It used to be a plagiarism-similarity check and little else. It now routinely includes image screening, reference verification, and increasingly checks for AI-generated text and figures.
Volume is what drives the change. More than 10,000 research papers were retracted in 2023, a record annual figure, according to Nature. Every retraction represents a screening opportunity that was missed years earlier, usually at a desk exactly like this one, and editorial offices have absorbed the lesson that catching a problem before review is cheaper than correcting the record afterwards.
What does the integrity layer look for?
Integrity screening at the desk targets the failure classes that are checkable without expert judgment about the science itself.
Images. Screening looks for duplicated panels within the manuscript, regions that overlap between figures, signs of splicing or local alteration, and figures reused from the published literature. Between 20 and 35% of manuscripts checked at acceptance are flagged for image-related problems, while acceptance is ultimately rescinded for only 1 to 8% of them, according to image integrity analyst Jana Christopher speaking to UKRIO. That ratio is the single most useful thing an editor can know about image flags, because it sets the correct tone for the query that follows.
Citations. Screening resolves each entry in the bibliography against the scholarly record to catch three distinct failures: references that do not exist, references whose metadata does not match the real work, and references to papers that have since been retracted. The fabrication problem has grown sharply with generative tools. A peer-reviewed test published in Scientific Reports found that a majority of citations generated by GPT-3.5 were fabricated, that GPT-4 still produced a substantial share of fabricated or erroneous citations, and that even citations to real works frequently contained substantive errors.
Statistics. Automated recomputation catches reported values that are internally inconsistent or mathematically impossible given the stated sample sizes, before a reviewer spends an evening on the analysis.
Declarations. Required statements are checked for presence and completeness, which is unglamorous, high-volume work that editorial assistants have historically done by hand on every single submission.
How does shared infrastructure change the picture?
Journals increasingly screen collectively rather than individually, and that is a structural response to a structural problem. Major publishers collaborate through the STM Integrity Hub on shared infrastructure for screening submissions, including paper-mill signals and other research-integrity concerns.
The reason a single journal cannot solve this alone is that the most damaging patterns are only visible across venues. A paper mill submitting fabricated manuscripts sends them to many journals; each journal sees one plausible-looking submission, while the network sees the same template arriving repeatedly. Duplicate submission works the same way: an author submitting simultaneously to three journals is invisible to each of them separately.
Shared screening also standardises what a signal means. When journals ran entirely separate processes, a flag at one venue carried no information for another, and an author whose manuscript was returned for an image query could simply submit elsewhere. Collective infrastructure changes that arithmetic without any single journal having to act as an enforcement body.
The remaining work stays local, and deliberately so. Shared systems surface signals; individual journals decide what those signals mean for a specific submission in their own field. A duplication pattern that is routine in one discipline may be unacceptable in another, and no central system should be making that call.
What happens when screening flags something?
A flag starts a process, not a punishment. COPE guidance gives editors a documented route for handling concerns raised about a published article, authorship disputes, and post-publication corrections, and a well-run editorial office follows that route from the first query onward.
In practice the first step is almost always a request for source material: the original uncropped image files, the underlying data table, or the specific reference in question. Most queries close at this step, because most flags have ordinary explanations and authors can supply what is asked for.
Two habits separate offices that handle this well from those that do not. The first is neutral language: the query describes what was observed and what is being requested, and it does not characterise the author. The second is recording the exchange, because a question raised about the paper years later needs the contemporaneous record of what was checked and what the author said. Integrity teams that later have to assemble evidence for a committee depend entirely on records created at this moment.
Escalation happens when the response does not resolve the question, and at that point the matter usually leaves the editorial office for the publisher's integrity team or the author's institution, which are the bodies with standing to investigate. Editors weighing an image flag should know the base rate first, because most flagged papers keep their acceptance rather than being withdrawn.
How Octym helps
Octym screens submissions across images, references, statistics, and required declarations, and returns the major risks first, each traced to its source and located in the manuscript. The review is configured per journal, so what runs and what leads reflects that journal's own instructions for authors. Findings are signals for editors to weigh, never verdicts about authors, and the decision stays with the desk. You can see how this fits an editorial workflow on the publishers page.