Oleafly Pre Submission

Pre-submission check

Everything that goes wrong at upload time is findable beforehand. This runs the checks and writes down what passed, what failed, and what only the author can decide.

Output is submission/checklist.md with a verdict per item. You do not fix things silently. You find them, report them, and fix only what the user asks you to fix.

1. Establish the target

Ask, or state your assumption:

• arXiv, a journal, a conference, or more than one?
• Initial submission, revision, or camera-ready? The rules differ at each stage.
• Double-blind, single-blind, or open?
• Which year and track?

The rules that matter are on the venue's current call for papers. Anything bundled here or in venue-templates is a snapshot that may be stale. Never state a page limit or a deadline from memory. Say where the number came from, or ask the user to confirm it.

When the venue-templates skill is enabled and read_skill_file is available, load_skill it and pull what applies:

• references/conferences_formatting.md
• references/journals_formatting.md
• references/reviewer_expectations.md

Then use references/venue-rules.md in this skill for the checks that apply regardless of venue.

2. Build clean

1. compile. Zero errors is the bar. A warning-free build is the goal.
2. get_log and scan for the warnings that matter. references/log-warnings.md lists them with what each means and whether it blocks.
3. project_map and read unresolvedRefs and unresolvedCites. Both must be empty. A ?? in the PDF is the single most common thing an author ships by accident.
4. get_pdf_text and read the rendered text end to end. Look for ??, [?], a missing figure box, a truncated table, an orphaned heading, or a bibliography that stopped early.

Record the page count and compare it against the venue limit. Count what the venue counts: some exclude references, some exclude the appendix, some count everything.

3. Source hygiene for arXiv

Only for arXiv, and only for LaTeX sources. Read references/arxiv-checks.md in full and work through it. The short version, each of which is a checklist row:

• The source builds with the engine arXiv will use. arXiv runs pdfLaTeX by default; Oleafly compiles with Tectonic, which is XeTeX. A document needing fontspec will not build on arXiv's default path.
• \pdfoutput=1 within the first few lines of the main file when the document is meant to be built as PDF output.
• The .bbl file is included. arXiv does not run BibTeX or Biber for you.
• No absolute paths in \input, \include, \includegraphics, or \bibliography.
• File names are plain ASCII with no spaces, no leading dots, and no two files differing only by case.
• Figures are PDF, PNG, or JPEG. No .bmp, no .tif, and no mixing an .eps and a .pdf of the same figure.
• Fonts are embedded, and none of them are Type 3 bitmaps.
• Supplementary material that is not compiled sits in anc/.
• The abstract for the metadata field fits arXiv's character limit. Check it before you open the form.
• The license choice is deliberate. It cannot be changed after announcement.

Typst projects

arXiv does not build Typst. It refuses a PDF that was made from TeX, but a Typst PDF was not, so a PDF-only upload is the normal route. Skip the LaTeX rows above and check these instead, for arXiv and for any other venue:

• The Typst version is on record. Read typst.version in project.json and write it into the checklist, so a camera-ready rebuild uses the same version. A project with no pin compiles with whatever default the app has at the time.
• Every @preview import names an exact version, and the project has compiled once while online so the packages are cached.
• The log has no unknown font family warning. Typst embeds and subsets every font it uses, so the risk is a silent fallback font, not a missing one.
• The PDF title and author come from #set document(title: ..., author: ...). Check that the title matches, and for a double-blind venue that author is empty or absent.
• A missing label or citation key stops a Typst compile with an error instead of printing ??, so a clean build already covers that row. Still read the bibliography in get_pdf_text, because an entry with missing fields renders without complaint.
• When the venue wants source files, the bundle holds the main .typ file, every file it includes or imports, the .bib or .yml bibliography, every image, and a note with the Typst version.

4. Front matter and metadata

• Title matches between the manuscript, the metadata, and any cover letter.
• Abstract present, within the venue's word limit, and free of undefined macros, citations, and references to figures.
• Author list complete, in the agreed order, with affiliations and any required identifiers.
• Corresponding author and contact details where required.
• Keywords or subject categories chosen.
• Funding, acknowledgements, data availability, code availability, ethics, and conflict statements present where the venue asks for them.

For a double-blind venue, all of the following must be absent from the submission: author names, affiliations, acknowledgements, funding statements naming a grant holder, a repository URL that identifies the group, self-citations written in the first person ("in our previous work [12]"), and identifying metadata in the PDF. In a Typst project that metadata comes from the author in #set document(...). Search for them with search_project rather than trusting a read-through. references/venue-rules.md has the search patterns.

5. Figures and tables

• Every figure and table referenced from the text.
• Raster figures at the resolution the venue asks for, usually 300 dpi for photographs and 600 or more for line art. Vector where the figure is a plot or a diagram.
• Text inside figures readable at print size, and no smaller than the caption text.
• Colour is not the only encoding, and the palette works for colour-blind readers.
• Captions self-contained: what is shown, what n is, what error bars mean.
• Nothing important sitting outside the trim area or in the gutter.

Use verify_pdf_pages to look at the rendered pages when PDF page capture is on in Settings. It rasterizes up to six pages, so pick the pages with figures. When it is off, fall back to get_pdf_text and list_files.