One .docx · one target · a few seconds
Twenty-eight reasons your paper comes back unread.
Before an editor reads one sentence, the editorial office runs a mechanical compliance pass. foliom computes those checks first — and proves it did not change a word of your paper while looking.
The same 5,592-word paper, per target
11 pages here. 26 pages there.
10 of the 12 targets. The rest are document classes their publishers distribute themselves, and nobody has typeset this paper into them yet, so they are absent here rather than guessed at.
targets, on 8 rows of the table below — counted from the registry the converter itself reads, as this page renders. One of those rows is not a publisher’s.
checks in one real IEEE-conference run: 19 passed, 5 failed, 5 warnings, 1 not applicable. Every one of them named, and reported whether it passed or not.
source paragraphs in that manuscript, each accounted for exactly once before anything was typeset.
document classes carry those targets — IEEE’s two profiles share one class file, ACM’s share another — and the ones a publisher distributes themselves are looked for on this server’s own disk as this page renders.
The mechanism, not a promise
The real question about a converter is not whether it works. It is whether it will change your paper.
A deterministic set of rules classifies every paragraph. Before anything is rendered, the structure that came out is checked against the document that went in: every source paragraph accounted for exactly once — nothing missing, nothing duplicated, nothing that was not in the file you uploaded. A conversion that cannot show that reports it as a failure against itself, in the same report as everything else, rather than handing you a PDF and staying quiet.
A skip is a result
A blinding check that passes because it was blind is worse than no check.
When foliom cannot run a check — a figure in a vector format has no resolution to measure, a venue that does not review double-anonymous has nothing to anonymise — it records a skip and says which check and why. It never quietly becomes a pass. No score, no grade, no ready-to-submit — it cannot know that.
Four passes over one file
What happens between the upload and the PDF.
-
Read the .docx
A .docx is a zip file, and this reads it as one: the document, its styles, its numbering, its equations, and every image with the pixel dimensions it was saved at. Nothing in the file is executed and nothing is fetched from it.
stops the job if: the upload is not a Word .docx this reader can open
-
Recover the structure
Rules classify every paragraph — title, author, abstract, heading, caption, reference. Only the paragraphs the rules cannot settle go to the second opinion below, and the recovered structure is then checked back against the source.
stops the job if: the document’s body cannot be read at all
-
Typeset it
The structure is written into the publisher’s own class file and compiled with XeTeX, or pdfTeX where the class requires it. The .tex it compiled comes back with the PDF, so nothing is locked in.
stops the job if: LaTeX produces no PDF — and the log comes back with the job
-
Check the result
The pre-flight report reads the finished PDF rather than your intentions: the page count from the PDF, the blinding from the PDF’s own text and file properties, the figure resolution from the images at the size they will print.
stops the job if: never — a failed check is a finding, not a crash
One check, expanded
Every finding says what it found, what the publisher wants, and what to change.
fig3 is 141 dpi (494 px wide), needs 600 dpi (2100 px)
IEEE requires 300 dpi for photographs, 400 for combination art and 600 for line art, at the printed size
re-export the charts at the pixel width given above, or as PDF, which has no resolution to be too low
the image inside your .docx — its pixel width over the width it is placed at — against IEEE’s 3.5-inch column
What comes back
Four files, and nothing you cannot open elsewhere.
The finished PDF
Typeset in the journal’s own template, ready to read.
The LaTeX source
Yours to keep editing. Nothing is locked in.
The submission bundle
Figures renamed to the publisher’s convention, arranged the way the upload form asks for them.
The pre-flight report
And, where the venue reviews double-anonymous, a blinded copy checked against the PDF itself.
A bounded second opinion
The language model answers multiple choice, and nothing else.
The rules run first and settle most of the document on their own. Only the paragraphs they could not settle are asked about, each as one question with the paragraph before and after it for context, and the reply is constrained by a schema to a choice, a confidence and a one-clause reason. There is nowhere in that answer to put prose. Where the second opinion is switched off, or the service does not answer, the report says so rather than guessing.
And no invented references
A citation whose reference is missing is printed exactly as you wrote it, and reported.
No entry is made up to get the bibliography to compile. Where the reference lookups are switched on, what is sent is the reference itself — authors, title, year, DOI, all of it already published — and never a line of your own text. A reference the lookup cannot confirm is reproduced as the manuscript wrote it and listed in the report.
Every target, and the class file behind it
Twelve targets, on the publisher’s own class file wherever there is a publisher
The table has eight rows, one per name in the Publisher column. Rows marked not a publisher are the exception the heading allows for: foliom’s own generic two-column article, built from base LaTeX, for a paper with no venue yet.
| Publisher | Targets | Class file |
|---|---|---|
| ACM | Conference (sigconf), Journal (acmsmall) | acmart |
| Elsevier | Final two-column, Preprint | elsarticle |
| IEEE | Transactions / journal, Conference proceedings | IEEEtran |
| IOP Publishing | Journal article | iopjournal |
| MDPI | Journal article, submission form | Definitions/mdpi |
| Springer Nature | LNCS proceedings, Journal article | llncs · sn-jnl |
| Taylor & Francis | Journal article | interact |
| foliom not a publisher | Two-column preprint | article |
The list is generated from the registry the converter itself reads, and every publisher-distributed class file is looked for on this server’s own disk as the page renders, so neither the targets nor their marks can drift from what this deployment actually runs.
Published nowhere.
A conversion is written under a key of its own and handed back only to the person it belongs to; every page that could show one is kept out of search engines by both the response header and the page itself. Nothing expires on its own, and deleting a conversion erases every file it has — what is kept, and for how long.
No analytics.
No tracking pixel, no advertising, no third-party script and no third-party font on any page here. The content security policy on every response allows scripts and styles from this origin only, which is why the display face is served from this server rather than linked from a font host that would see every visit.
No publisher endorses this.
A class file’s name is a technical fact, not an endorsement. foliom is not affiliated with any publisher named on this page, and states which template it used so you can check the result against the author guide yourself.
Why it asks for an account
An upload is an unpublished manuscript.
It is stored against a person — someone who can sign in, see every file they have uploaded, and delete all of it — not against a browser cookie nobody can revoke. What is kept, and for how long, is written down.
Upload the paper you are reformatting this week
Find out what comes back before an editor does.
If the conversion cannot account for every paragraph of your source, the report says so instead of handing you a file that looks finished.