paperpush validate

Run the pre-submission checks on a filled .sub file. Run this before paperpush submit to catch problems while they are still cheap to fix.

usage: paperpush validate [-h] [-v] [-q] [--dont-check-links]
                          [--dont-check-for-sensitive-info]
                          [--dont-check-references]
                          [--dont-check-manuscript] [--anonymous]
                          [--dont-check-hidden-text]
                          [--dont-check-openreview-profiles]
                          [--arxiv-latex-cleaner]
                          subfile

Synopsis

paperpush validate biorxiv.sub

validate reports two severities:

• Errors must be fixed before the file can be submitted (for example, a required field left blank, a referenced file that does not exist, or a choice value that isn’t one of the venue’s allowed options).

• Warnings are worth reviewing but don’t block submission.

The command exits non-zero if any errors are found, so it fits cleanly into scripts and CI.

What it checks

This is a summary. What validate checks lists every check in detail, with its severity and the flag that controls it.

• Required fields are present and non-empty.

• Files exist — every file / filelist path (manuscript, figures, cover letter, supplements) resolves to a real file on disk.

• Choice values are valid — every choice field holds one of the venue’s allowed options, including nested drill-down fields.

• Structured fields are well-formed — for example, author lists parse and mark exactly one corresponding author.

• Cited links resolve — the URLs in the uploaded files are probed, and ones that are definitively broken (404/gone, including a still-private GitHub repository) are reported as warnings.

• Reference DOIs point at the right paper — every DOI in the submission’s bibliography is resolved through doi.org and held against the reference citing it. A DOI that is malformed, cited by two references, unregistered, or registered to a different work — the usual sign of a DOI copied from the wrong reference — is reported as a warning.

  The bibliography is read whichever way the submission ships it. If there is a .bib file (including one bundled in a source archive), its entries are compared field by field: title, first author, and year. Otherwise the manuscript’s own reference list is read — from a PDF’s extracted text, LaTeX source, or a compiled .bbl — and each DOI is compared against the text of the reference entry it sits in, so no .bib is needed. Only DOIs inside the reference list are compared this way; a dataset DOI in a data-availability sentence is merely checked that it resolves.

• Nothing private is about to be published — the uploaded files are scanned for API keys, passwords, private keys, GPS coordinates in figures, editable-document links, and LaTeX source comments.

• The manuscript meets the venue’s author guidelines — the uploads are measured against the rules recorded for the venue in manuscript_requirements.json (see paperpush requirements to print them):

  • the manuscript file’s format and size, and its word and page count against the venue’s limits, for the article type selected in the .sub. A .tex manuscript (or a .zip LaTeX bundle) is compiled to a scratch PDF with latexmk or pdflatex for its page count; the source directory is never written to, and without a TeX toolchain the page limit is reported as unchecked;

  • the section headings and declarations the venue requires (Introduction, Methods, Results, …; data availability, competing interests, author contributions, …), found in the manuscript text under any of their usual wordings;

  • title-page items that leave a trace in the front matter — the title, a corresponding e-mail, an ORCID, a keywords line, a running title, a word count;

  • abstract, title, running-title, and keyword limits read from the .sub;

  • each figure’s format, file size, resolution (from its metadata, or judged from its pixel width at the venue’s print width), colour mode, and print dimensions, and the number of figures, tables, and display items;

  • supplementary-file format, size, and single-PDF rules; the cover letter; the reference count; the total upload size.

  Measured numbers that exceed a limit are errors; anything that depends on extracting text from a PDF or reading an image is a warning. A rule the portal already enforces through venues.json (an accept list, a per-file size cap, a word limit on the field) is reported once, by that check, not again here.

• Nothing is hidden from the reader: white, invisible, microscopic, or off-page text in the PDF, LaTeX, or Word uploads. A hidden instruction to an AI reviewer (prompt injection) is an error.

• The template is unmodified: for venues with a recorded template layout (ICLR), the manuscript PDF’s margins, body font, line spacing, running head, and review line numbers are measured against the official template.

• Every author has a usable OpenReview profile: for OpenReview venues (ICLR, AAAI), using the login stored by paperpush login.

The link, reference, sensitive-information, manuscript, hidden-text, and OpenReview passes make network requests or read every uploaded file; each can be turned off with its --dont-check-* flag below.

Arguments

subfile

Path to the .sub file to check, e.g. biorxiv.sub.

--dont-check-links

Skip probing the URLs cited in the uploaded files. Avoids network requests.

--dont-check-references

Skip resolving the references’ DOIs through doi.org. Avoids network requests.

--dont-check-for-sensitive-info

Skip scanning the uploaded files for information not meant to be published.

--dont-check-manuscript

Skip measuring the uploads against the venue’s author guidelines. Also skips compiling a LaTeX manuscript for its page count.

--dont-check-hidden-text

Skip looking for white, invisible, microscopic, or off-page text. See Hidden text and prompt injection.

--dont-check-openreview-profiles

Skip looking up the authors’ OpenReview profiles (OpenReview venues only). Avoids network requests. See OpenReview author profiles.

--anonymous

Check the submission for information that identifies its authors, as for a double-blind venue. Venues marked anonymous in venues.json (ICLR, AAAI) get this check automatically. Linked anonymous.4open.science repositories are fetched and scanned too, which needs network access. See Anonymization (double-blind venues).

--arxiv-latex-cleaner

Run arxiv_latex_cleaner on the LaTeX source first, then validate the cleaned copies. Applies to every .tex upload (the cleaner runs on its whole directory) and every .zip, .tar, .tar.gz, or .tgz bundle that contains a .tex file. The cleaner strips comments and drops unreferenced and auxiliary files. Cleaned copies are written next to the originals, as paper/ → paper_arXiv/ and source.zip → source_arXiv.zip, replacing any earlier output. The originals and the .sub are not changed, so point the .sub at the cleaned copies to submit them. Needs the optional package: pip install "paperpush[validate]".

Plus the common -v/--verbose and -q/--quiet logging flags. Use -v to see the checks as they run.

Examples

paperpush validate biorxiv.sub

Fail a CI job if the submission isn’t ready:

paperpush validate biorxiv.sub || exit 1

Run the field and file checks alone, with no network access:

paperpush validate biorxiv.sub --dont-check-links --dont-check-references

Clean arXiv LaTeX source and check the result:

paperpush validate arxiv.sub --arxiv-latex-cleaner

See also

• What validate checks — every check in detail.

• paperpush options — look up the valid values for a field flagged as invalid.

• paperpush requirements — print the author-guideline rules the manuscript is measured against.

• paperpush submit — the next step once validation passes.