scripts

scripts

Maintenance scripts for the QPL 2026 program booklet.

verify_abstracts.py

Checks every abstract printed in assets/pdfs/QPL_2026_Program.pdf against its original, and repairs the ones that have drifted.

The booklet is typeset from assets/pdfs/abstracts.typ, whose bodies were lifted from the front matter of the submitted papers in assets/submissions/ and — where a submission carries no abstract — from the paper’s arXiv version. The script re-derives both originals and compares. An abstract that matches either one is left alone; one that matches neither is replaced with an exact copy of the original, preferring the submitted paper and falling back to arXiv. The “Abstract from the arXiv version, arXiv:…” footnote is kept in step with the body it labels.

Comparison ignores what a PDF extractor can legitimately change — ligatures, hyphenation across line breaks, quote and dash shapes, $…$ delimiters, case and whitespace — so only a real difference in wording counts as a mismatch.

pip install pymupdf

./scripts/verify_abstracts.py            # report only; exit 1 if anything would change
./scripts/verify_abstracts.py --diff     # report + word diff for each mismatch
./scripts/verify_abstracts.py -v         # also list the abstracts that check out
./scripts/verify_abstracts.py --apply    # rewrite abstracts.typ

cd assets/pdfs && typst compile QPL_2026_Program.typ   # after --apply

Talks are matched to their submission through the schedule tables in program.md, so re-scheduling a talk or swapping its PDF link needs no change here.

Outcomes

state	meaning
ok	matches the submission or the arXiv version
mismatch	matches neither — rewritten with the original
relabel	body is right but the arXiv footnote is missing, stale, or spurious
now-available	printed as “[No abstract]” but a source has since appeared — filled in
still-missing	no abstract in the submission and no arXiv version known
no-original	nothing to compare against, so the printed text stands
pinned	listed in abstract-pins.json; never touched
manual	body splices in a formatted Typst variable; reported, never rewritten
unmatched	no talk in program.md with that title and time slot

Data files

• arxiv-ids.json — submission PDF → arXiv id. Seeded from the original build (fuzzy title/author matches, checked by hand). --discover-arxiv adds ids printed on the papers themselves; anything but arXiv’s own stamp must agree with the talk’s title or author list before it is trusted, so a preprint merely cited in the front matter is not mistaken for the paper.
• arxiv-abstracts.json — cache of fetched arXiv abstracts, so a re-check needs no network. --refresh-arxiv re-fetches.
• abstract-pins.json — talks whose printed abstract is authoritative even though it matches no source, e.g. text supplied directly by the authors.

gen_posters.py

Writes assets/pdfs/posters.typ, the poster abstracts that close the booklet. Talks and posters are kept in separate files because they come from different places: a talk’s abstract is read off the submitted paper, while a poster’s is the one its authors typed into EasyChair. verify_abstracts.py therefore has nothing to say about posters, and does not look at this file.

Three inputs, none of which duplicates the others:

• the EasyChair abstract export (plain text) — the abstract of every submission, as its authors wrote it;
• the board-assignment spreadsheet (.xlsx) — which submission hangs on which board, i.e. which posters are actually being presented and in what order a visitor walks past them. Only submissions listed here are printed, and a submission spanning several boards is printed once, over the range;
• accepted.md — the canonical title and author list for each poster, matched on title, so that the booklet and the website agree. A poster the spreadsheet names but the page does not list is reported and skipped.

./scripts/gen_posters.py ~/Downloads/'qpl2026-abstracts posters.txt' \
                         ~/Downloads/'Indeling posters.xlsx'
cd assets/pdfs && typst compile QPL_2026_Program.typ

Unlike the talk abstracts, these arrive as LaTeX source rather than as typeset text, so the markup has to be resolved before it can be set: \emph and friends are unwrapped, TeX quoting and dashes become the characters they stand for, and each maths span is rewritten into Typst maths. A span spelled in a way the rewriter does not know raises rather than reaching the page as visible markup — extend MATH_SUBS when that happens. A closing paragraph that only links to the full version is moved out of the body and set as a footnote; the handful that read awkwardly there are reworded in NOTES.

abstract_extract.py

The PDF front-matter reader used by verify_abstracts.py. The submissions are author-formatted with no common template, so the abstract is found geometrically: an explicit “Abstract” label first, and failing that the first prose block on page 1 that is set apart from the body text (indented, narrower, smaller, or italic) and sits above the first section heading. It also reads the arXiv id off a paper when there is one.