Weekly publication sync

Keeps _pages/publications.md in step with Google Scholar / arXiv. Run by a weekly scheduled task; can also be run by hand.

1. Find differences

python3 scripts/pubsync.py

• new_preprints: arXiv records not yet on the page.
• now_published: entries under Preprints whose arXiv record now has a journal ref or DOI.
• still_preprint: every arXiv ID still under Preprints. arXiv’s journal fields are often never updated by authors, so also look each title up on Crossref (https://api.crossref.org/works?query.bibliographic=<title>&rows=3) and in any Google Scholar alert emails (sender scholaralerts-noreply@google.com). Only treat a paper as published when the DOI’s title and author list match.

Google Scholar itself blocks automated access, so it is not queried directly.

2. New preprint → new row at the top of ## Preprints

Row template (one line, newest first):

| <img src="/assets/images/<IMAGE>" width="200" alt="" align="center" style="display:block;margin-bottom:10px;margin-top:10px;margin-left:0px;margin-right:0px;padding-left:5px;padding-right:5px;" /> | **<Title in sentence case>** <br> _with <co-authors, Oxford comma, Brendan omitted>_ <br> arXiv: <ID> [<primary category>] [[preprint](https://arxiv.org/abs/<ID>)] <a href="#bibtex-<KEY>" class="bibtex-toggle" data-bibtex="<KEY>">[<span>BibTeX</span>]</a> |

Add the matching entry to _data/bibtex.yml (see section 4).

Choosing the image:

1. Download the source: https://arxiv.org/src/<ID> (a gzipped tarball, occasionally a bare gzipped .tex).
2. Render every figure (pdftoppm -png -r 80), look at them, and prefer, in order: a 2D/3D solution or geometry plot (colourful, readable at 200 px) → a single-panel convergence/comparison plot with a legend → one panel cropped from a multi-panel figure. Avoid schematics that are mostly text, tables, and multi-panel grids at full size.
3. Crop to the content (convert -trim), resize to at most 600 px wide, strip metadata, and save as assets/images/<short_descriptive_name>.png.
4. If the source has no usable figure, use paper.gif and the grey-background style used by the “isometry constraint” row.

3. Published preprint → move to ## Peer-Reviewed Scientific Articles

Remove the row from Preprints and insert it in Peer-Reviewed, ordered by year (to-appear entries first). Keep the image cell. Replace the venue line with:

<Full journal name> (<year>) [[preprint](https://arxiv.org/abs/<ID>)] [[journal doi](https://doi.org/<DOI>)]

Keep the [BibTeX] toggle at the end of the cell, and replace the paper’s @misc entry in _data/bibtex.yml with an @article built from the Crossref record (section 4). Keep the same key unless the year changed; if it did, rename the key in both files.

The year is that of the printed volume; for online-first papers with no volume yet, use the online year and update it once the volume appears.

Accepted-but-not-yet-online papers use <Journal> (to appear, <year>) and get the DOI later. Keep any existing [[code]] links.

4. BibTeX entries

Every row ends with a [BibTeX] toggle whose data-bibtex key names an entry in _data/bibtex.yml (a YAML block scalar per key). Build entries only from the arXiv API record or the Crossref record (https://api.crossref.org/works/<DOI>), never from memory.

• Key: <first author surname><year><first significant title word>, lowercase ASCII (e.g. fu2026locally); on a clash pick a later distinctive word.
• Preprint: @misc with title, author, year, eprint, archivePrefix={arXiv}, primaryClass, doi={10.48550/arXiv.<ID>}, url.
• Journal article: @article with title, author, journal, volume, number, pages (-- ranges; article number if no range), year (volume/print year), doi, url.
• Authors as Family, Given and ... with full given names (fill initials from arXiv); accents as LaTeX (J{\o}rgen, G{\"u}nther); brace acronyms and proper nouns in titles ({Galerkin}, {SiMPL}, {PDE}).
• Check: every data-bibtex key on the page exists in the YAML, and the entries compile (\nocite{*} + bibtex) without warnings.

5. Deliver

Open a pull request against master titled Publications sync <YYYY-MM-DD> listing each change and, for every new row, which figure was chosen and the runner-up. If nothing changed, do nothing.