Figure version control is a naming and record discipline: every figure file carries its number, version and date; a short log records the source data file, the generating script and what changed. Update the log at each revision round, then check the log against the proof so the published figure matches the accepted text.
What figure version control is, and where a stale figure gets in
A figure version is one rendering of one figure, produced from one state of the data by one state of the code. Change the data or the code and you have a new version, even if the picture looks the same. Figure version control is the practice of giving each rendering a distinct, sortable identity and recording what produced it.
A revision round that changes the data or the analysis produces a new rendering of a figure. These are the routes by which a stale one reaches production:
- Two files with the same name in different folders, and the older one is attached to the email.
- A figure regenerated from corrected data while the caption and the results paragraph still quote the earlier numbers.
- A panel reordered in the manuscript but not renamed on disk, so Figure 3 is uploaded as figure 2.
- A co-author returning an edited figure with no record of which version they started from.
- A low-resolution screenshot pasted in for review and never swapped for the print-resolution file.
None of these is a failure of care; each is a failure of identification. If a file cannot tell you which version it is, the proof will not catch the substitution.
A figure file-naming convention that survives three revision rounds
Use one pattern for every figure file, from first draft to accepted version. The pattern below sorts correctly in any file browser and reads aloud without ambiguity.
fig02_v03_2026-03-06_survival-curve.tiff
| Segment | Example | Rule |
|---|---|---|
| Figure number | fig02 | Zero-padded to two digits, and identical to the number used in the manuscript text |
| Version | v03 | Increment on every regeneration. Never overwrite an earlier version |
| Date | 2026-03-06 | ISO 8601, the date the file was generated, not the date it was sent |
| Short label | survival-curve | Two or three lowercase hyphenated words naming what the figure shows |
| Extension | .tiff | The file format the journal asks for, at the resolution it asks for |
Four rules make the convention hold under pressure:
- No spaces and no capitals. Spaces break scripts and are silently rewritten by some mail systems.
- The words final, latest, new and revised are banned. They describe a moment rather than a version, and the moment moves.
- If figures are reordered, rename every affected file in the same sitting and record the renumbering. Never leave the number in the filename disagreeing with the number in the text.
- Keep a copy of exactly what you sent in a folder named for the round, such as submitted/round-2/, and treat it as read-only from the moment the email leaves.
The figure log: one table that records every version you made
A figure log is a single table, kept with the manuscript, in which one row records one version of one figure. Keep it in plain text or a spreadsheet rather than inside the manuscript file, so it is never accidentally submitted.
| Figure | Version | Date | Source data file | Generating script | What changed in this version |
|---|---|---|---|---|---|
| Fig 1 | v01 | 2026-01-12 | screening_counts_2025-12-20.csv | fig01_prisma_flow.R | First submission version |
| Fig 1 | v02 | 2026-03-04 | screening_counts_2026-02-28.csv | fig01_prisma_flow.R | Search rerun to 28 February at reviewer 2 request. 41 more records identified, 3 more included. Caption counts updated to match |
| Fig 2 | v03 | 2026-03-06 | outcomes_clean_2026-03-05.csv | fig02_survival.R | Two participants removed after a data query. Follow-up censored at 24 months, not 36. Results paragraph 3 updated |
| Fig 3 | v02 | 2026-03-07 | Not applicable, photomicrograph | fig03_crop_notes.md | Scale bar added and panel relabelled B. No brightness, contrast or colour adjustment applied |
The last column earns its keep. Write it for a reader who is not you, six months from now: say what changed in the data or the code, and name the parts of the manuscript you changed to match. That second habit stops a corrected figure sitting above an uncorrected sentence.
What to record when a figure has no script
Not every figure comes from code. A photomicrograph or a clinical photograph is prepared in an image editor, so the generating script column holds the name of a short notes file instead. Record every adjustment you applied, in order, and say explicitly where you applied none. The boundary between legitimate adjustment and manipulation is set out in image integrity in research, and the log is where you show which side of it you were on.
Keep the figure, its source data and its script in one folder
Version chaos starts when the three things that make a figure live in three places: the data on a shared drive, the script in a notebook, the image in a downloads folder. Keep them together:
figures/source-data/— the exact data file each figure was drawn from, dated in the filename, never edited in place.figures/scripts/— one script per figure, named for that figure, with the input filename written inside it rather than typed at a prompt.figures/output/— every rendered version, named fig02_v03_2026-03-06_survival-curve.tiff: figure number, version, date, short label.figures/submitted/round-1/andfigures/submitted/round-2/— frozen copies of what actually went to the journal.figures/figure-log.md— the log table itself.
If you use Git, commit the scripts and the log and leave the large images out of it. If you do not, a dated folder copy at the end of each round achieves most of the same protection. The point is that a figure, its input and its recipe can always be reunited.
What to check against the figure log when the proof arrives
The proof is the last moment at which a figure can be changed without a public record. Work from the log rather than from memory:
- For each figure, read the version number from the frozen round folder and confirm the proof shows that rendering.
- Read every caption against its log row. Numbers quoted in a caption, such as the counts in a study flow diagram, must match the version printed above them.
- Read the results text and the abstract against the same log rows. A change in the data behind a figure is a change in the sentences that report it.
- Check panel letters and figure order against the numbering in the text, and against every callout such as see Figure 2.
- Confirm resolution, orientation and that the whole figure is present. A cropped or rotated panel is a production error only you can see.
- Confirm supplementary figures still carry their own numbering series and have not been merged into the main sequence, as covered in preparing supplementary materials.
A flow diagram deserves its own pass. A PRISMA or CONSORT diagram, drawn from templates published by the PRISMA and CONSORT groups and catalogued in the EQUATOR Network library, is built from counts that change whenever a search is rerun or a participant is reclassified. The ICMJE Recommendations ask that figures be self-explanatory and numbered consecutively in the order they are cited in the text; the counts inside the boxes belong to the reporting guideline, so a diagram that disagrees with the Methods fails that guideline even when the image itself is clean.
Directive Publications does not require a figure log, and no published policy of ours asks for one. It is a private discipline that protects you, not a submission requirement. What we do ask for is set out in our author guidelines: figures numbered and cited in the text, each with a concise caption, supplied at high resolution — at least 300 dpi — in TIFF, EPS, PDF or high-quality PNG or JPG, and either embedded in the manuscript file or supplied separately.
If the wrong figure version has already been published
Tell the editorial office as soon as you find it, and say plainly which version was published and which should have been. A figure error that changes what the article reports is corrected through the published record, not by quietly replacing a file: the version of record is what others have already read, cited and downloaded. The Committee on Publication Ethics treats correcting the literature as a core practice, and a correction notice given its own DOI can carry Crossref metadata that ties it to the original article, so the two travel together.
A figure log makes that conversation short. You can state which version was sent, on what date and from which data file, and what the correct version shows instead. Without one, the exchange becomes an argument about attachments.
Before your next revision goes out, read the figure requirements on Author Guidelines and our Corrections Policy, so you know what can still be fixed and what becomes a public notice. Related reading: preparing figures and tables for publication for format and resolution, and how to respond to reviewer comments for the revision round these files belong to.