Home / Guides

Pre-submission checklist: 12 consistency checks before you submit a paper

The mistakes that survive careful proofreading, and how to catch them in the last read before you submit.

DeskReady · October 2026

Proofreading catches typos. It rarely catches the other kind of mistake: a number in the abstract that no longer matches the results table, a total that doesn't add up, a caption that describes a different figure. These slip through because each sentence looks fine on its own. The problem only shows when you hold two places in the paper side by side.

Reviewers and editors do exactly that, and an inconsistency in the first few pages makes them read everything after it more sceptically. Below are twelve checks for the final read. The examples come from 65 published arXiv preprints we ran through DeskReady; several of those papers later appeared at major conferences. Papers are not named.

Numbers

1. Every number that appears twice agrees with itself. Results tend to be quoted in the abstract, the introduction, the contributions list, the results section and the conclusion. When an experiment is re-run, the results section gets updated and the others often don't.

   From a machine-learning paper

   The contributions list reports peak memory savings of 22.8% and 14.9%. The results section, measuring the same models against the same baseline, reports 23.8% and 15.6%.

   Fix: use the results-section numbers in the contributions, where most readers will see them.

2. Totals and percentages add up. Recompute every “in total”, every sum of subgroup counts, every set of shares that should reach 100%.

   From a medical-imaging paper

   The text names four organs, then gives counts for three of them and “a total of 281”. The three counts already sum to 321 (131 + 64 + 126). The fourth count was missing, and 281 was in fact the size of that missing set.

   Fix: add the missing count and recompute the total, or drop the word “total”.

3. Arithmetic inside figures and tables is right. Figure labels and table cells are made in a different tool than the text and are easy to copy forward from an older version.

   From an economics working paper

   A figure is labelled “Simulations Run: 2 × 5 × 8 = 405”. Two pages earlier the prose has it right: 2 × 5 × 8 = 80. The figure label had been copied from an earlier figure.

   Fix: change the label to 2 × 5 × 8 = 80.

4. Rounding is consistent. A value rounded to 5.3 in one place and quoted as 5.31 in another, or “less than 1%” for something that is 1.03%, reads as carelessness even when the result is sound.

Claims and results

5. Every sentence that describes a table matches the table. Look hardest at words like “all”, “every”, “consistently” and “significantly”.

   From a computer-vision paper

   The text says the method “significantly outperforms” the baseline “in every category”. In the table it cites, the baseline wins one category, 0.91 to 0.85.

   Fix: say what the table shows: better in 9 of 10 categories.

6. The settings in the method section are the ones the results used. Defaults described early in a paper are often superseded by an ablation later on.

   From a computer-vision paper

   The method section says a key parameter “is set to t = 0 by default”. The ablation five pages later finds the best performance at t = 200, and the main results use t = 200.

   Fix: state the value the reported results actually use.

7. Everything promised is delivered. Go through the abstract and the contributions list and find, for each claim, the experiment that supports it. Datasets, settings or ablations announced up front sometimes never appear.

Figures, tables and cross-references

8. Captions describe the figure they sit under. Compare each caption with the labels, legend and panel letters of its figure.

   From an astrophysics paper

   A figure caption says inset (a) shows “HH 201”. The label on the figure and the text that discusses it both say HH205, a different object.

   Fix: change the caption to HH205.

9. Every reference to a figure, table or equation points to the right one. “As shown in Table 3” should lead to a table that shows it. Check that nothing renders as “??” after the final compile.
10. Terms and symbols mean one thing throughout. The same quantity should keep the same name and symbol in the text, the equations and the figure legends.

Text and references

11. Nothing is said twice. Drafts merged from several authors, or rewritten over months, leave pairs of sentences that make the same point in different words. Reviewers notice the padding.
12. The reference list is clean. Look for duplicate entries, the same author spelled differently in the text and the bibliography, and broken entries such as stray braces from BibTeX.

How to run the check yourself

• Read the typeset PDF, not the source. Many of these problems only exist after compilation: figures from another file, tables pasted as images, numbers filled in by macros.
• Make a ledger of your key numbers. For each headline result, search the PDF for it and write down every page it appears on. Mismatches jump out.
• Read each figure and table next to the paragraph that cites it, not in sequence.
• Do one pass after the last experiment, not before. Most of the examples above are what the paper looked like after a re-run updated one place and not the others.
• Ask someone who hasn't read the draft. Co-authors who wrote a section read what they meant, not what is on the page.

What this checklist doesn't cover

It is about internal consistency: whether the paper agrees with itself. It doesn't replace the target journal's author guidelines (formatting, word limits, statements on data, ethics and conflicts of interest), and it doesn't judge whether the science is sound or the venue is right. Those still need the journal's own checklist and your co-authors.

Two checks get a guide of their own: numbers that don't add up and text that disagrees with its tables.