Working paper series tooling

cover.py writes the working paper cover page, either as LaTeX for a paper’s own source or stamped onto a built PDF, and series.json holds the series metadata that every cover is drawn from.

How the published PDFs are built

Each paper lives in its own repository and carries the cover in its own source, so a build of the paper produces the cover with it:

• the four LaTeX papers hold it as a titlepage before \maketitle, generated with --emit-latex
• paper-skills-dna holds it as paper/cover.tex, generated with --emit-latex --inline-abstract --pandoc and passed to pandoc with -H

The PDFs in assets/papers/ are those builds. A clean checkout cannot yet rebuild every paper, because the figures and generated tables of some are not committed and the microdata behind them is not either. For the September 2026 builds, a figure missing from a checkout was taken unchanged from the previously published PDF, so every figure is the one already published. In the skills-dna paper the recovered third figure is cropped to remove a line of body text that had been captured with it, and text from the surrounding page that sat invisibly inside the recovered figures is removed.

Without --emit-latex, cover.py stamps the cover onto a PDF as it was built, which remains the route for a paper whose source cannot be built. Nothing inside the paper is touched in that case, and the stamped file is the built paper with one page in front of it.

Running it

python tools/cover.py --all
python tools/cover.py --slug flows-engine
python tools/cover.py --slug flows-engine --force

A paper that already carries a cover is skipped, so running --all twice does not stamp a cover onto a cover. --force replaces the existing one. Where a paper’s own repository tracks a copy of its PDF, the stamped file is mirrored there too, so the published file and the repository agree.

The tool needs pypdf and reportlab, and it expects the sibling paper repositories to sit beside this one.

The series

Numbering is AC-WP-<year of first draft>-<sequence within that year>, and the sequence runs in order of first draft. Three numbers are reserved in series.json for papers whose estimation is not yet run, so that nothing renumbers when they are published.

The year in the identifier and in the suggested citation is the year of the first draft, and a revision keeps it.

series.json also holds each paper’s keywords and JEL codes, which the cover prints under the abstract and which are written into the PDF metadata, so they can be pasted straight into an SSRN or MPRA submission. Keywords use American spelling, because that is how both services are searched, although the papers themselves are written in British spelling. Each paper prints the same two lines under its own abstract, so a printed copy carries them whether or not it was taken from the cover.