The pre-submission manuscript checklist

Before submitting a manuscript, check six areas: figures and images, statistics, references, methods completeness against the relevant reporting guideline, ethics and disclosure statements including AI use, and the specific rules of your target journal and funder. Most desk rejections and integrity queries trace back to one of these areas, and every one of them is checkable before submission.

Are the figures and images clean?

Figures and images draw heavy scrutiny in editorial screening, so start the checklist there. Image integrity analyst Jana Christopher, in an expert interview with UKRIO, reported that 20 to 35% of the accepted manuscripts she screened before publication were flagged for image-related problems, and that one explanation she hears frequently is authors using placeholder panels during figure assembly and then forgetting to swap them before submission. That is exactly why checking your own figures pays off: errors made during figure assembly are removable before anyone else sees them.

• Every image appears once. No duplicated or overlapping panels within the figure set, and no reuse from your earlier papers without citation and permission.
• Adjustments are uniform and disclosed. Brightness, contrast, and color changes apply to the entire image and are declared where the journal requires it.
• Composites are marked. Spliced gel lanes and assembled panels are clearly delimited and described in the legend.
• Originals are archived. Uncropped, unadjusted source images are saved and ready to share on request.
• Legends match reality. Each legend describes what the figure actually shows, including magnification, staining, and replicate counts.

Image and figure problems are one dimension of a longer list worth screening; what a full integrity check covers is a useful map of the rest.

Do the statistics hold together?

Statistics are where honest errors concentrate, because numbers get transcribed by hand between analysis output, text, tables, and figures. Every transcription is a chance for drift, and reviewers notice drift.

• The test fits the design. The reported statistical test matches the data type, the sample structure, and the stated hypothesis.
• The numbers are internally consistent. Test statistics, degrees of freedom, and p-values agree with each other, and exact values are reported rather than bare thresholds.
• Sample sizes match everywhere. The n in the abstract equals the n in the methods, the tables, and the figure legends, and any attrition is explained.
• Percentages add up. Subgroup counts and percentages sum to their stated totals.
• Variability is defined. Error bars and spread measures are labeled consistently as SD, SEM, or confidence intervals.

The reason to be strict here is reproducibility. More than 70% of 1,576 researchers surveyed said they had tried and failed to reproduce another scientist's experiments, and more than half had failed to reproduce their own, in a survey published by Nature. A manuscript whose numbers cannot be recomputed from its own tables invites that failure, and statistical consistency is the part of reproducibility you fully control before submission.

Is every reference real, accurate, and still standing?

A bibliography fails in three distinct ways: a cited paper can be retracted after you cited it, it can contain errors that break the link between claim and source, or it can simply not exist. All three are checkable before submission.

• Every DOI resolves. And it resolves to the exact work you meant to cite, with matching authors, year, and title.
• No retracted papers remain. Retraction usually happens after citation, so a bibliography that was clean at drafting can be stale at submission. Re-check right before you submit.
• Each citation carries its claim. The cited source actually says what your sentence attaches to it. Citations copied from another paper's bibliography without reading the source propagate that paper's errors into yours.
• AI-assisted drafts are verified item by item. In a peer-reviewed test published in Scientific Reports, a majority of the bibliographic citations GPT-3.5 generated were fabricated, GPT-4 still produced a substantial share of fabricated or erroneous citations, and even citations to real works frequently contained substantive errors.

Fabricated references are convincing precisely because they are assembled from plausible author names, real journal titles, and reasonable page numbers. Nothing about how a reference looks tells you whether it exists; the check is resolving it against the actual record.

Are the methods complete enough to reproduce?

Methods completeness determines whether anyone, including a future version of you, can rerun the work, and reporting guidelines exist so that completeness is not guesswork. The EQUATOR Network maintains a comprehensive searchable library of reporting guidelines for health research, and many journals now ask for the matching checklist at submission. Four common study types and their guidelines:

Beyond the guideline checklist itself:

• Availability statements are accurate. Data and code availability statements are present, and the repositories they point to actually contain what they promise.
• Materials are identified precisely. Reagents, antibodies, cell lines, strains, and instruments are described well enough for someone else to obtain the same ones.
• The analysis path is complete. Software versions, parameter choices, exclusion criteria, and any deviations from a registered protocol are stated.

Find the guideline for your design before writing the methods section, not after; retrofitting a checklist onto finished text is where inconsistencies creep in.

Are the disclosures and ethics statements in place?

Disclosure and ethics statements are read before the science is, and gaps here delay manuscripts that are otherwise ready.

• Ethics approvals are stated with their approval numbers, and consent statements match what participants actually agreed to.
• Registrations are cited. Clinical trial and protocol registrations appear with their identifiers.
• Conflicts and funding are complete for every author, with grant numbers where the funder expects them.
• Author contributions are listed, and everyone named as an author qualifies as one.
• AI use is disclosed. The ICMJE Recommendations require authors to disclose AI-assisted technologies used in producing the submitted work, which covers data collection, analysis, and figure generation as well as writing.

The ICMJE position on authorship is equally clear: chatbots cannot be listed as authors, because authorship carries accountability for the whole work and a tool cannot accept accountability (ICMJE Recommendations). Disclosure is not an admission of anything. Undisclosed use discovered later is what turns a routine editorial question into an integrity question, so say what you used, where, and for what.

Does the manuscript meet the rules of where it is going?

The target journal's guidelines and the funder's requirements are the checklist areas most often left to the end, and they are pure compliance: no judgment calls, just rules that are public and specific.

• Author guidelines are followed to the letter: structure, word and figure limits, reference style, file formats, and the journal's data and code policies.
• The journal's integrity policies are read, not skimmed. Journals differ on image-adjustment rules, AI-use wording, and preprint handling, and the differences matter.
• Funder requirements are met. NIH (for R01 and R21 awards), ERC, Horizon Europe, Wellcome, Gates, HHMI, and ISF each publish their own expectations on open access, data management, and rigor, and a manuscript that satisfies the journal can still fall short of the grant's terms.
• Preprint status is declared if the journal asks, along with any prior submission history it requires.

These rules vary enough between venues that a generic pass is not the same as one configured to where the manuscript is going. This is also where a checklist becomes a habit: researchers and grant applicants who make this pass routine spend revision rounds on science rather than on formalities. Two checklist items have their own articles worth reading: verifying references a language model produced and disclosing AI use correctly.

How Octym helps

Octym runs this checklist as one review: figures, statistics, references, methods, and disclosures in a single pass across every dimension, on the Octym platform. Suspected issues surface as ranked signals with the evidence behind each one, traced to its source in the manuscript or the literature, so you can see what a screener might question before a screener does. You decide what to address; nothing here is a verdict.