Make sitemap lastmod accurate and stable

[LukasWallrich]

Closes #859

Root causes

• 740 URLs without lastmod: Hugo takes page dates from git log --name-only, and Git quotes paths with non-ASCII characters by default ("content/glossary/turkish/a\303\247\304\261k_bilim.md"). Hugo's parser never unquotes them, so no glossary term or curated resource with an accented or non-Latin filename got a date. Verified locally: setting core.quotepath=false takes the count from 738 to 1.
• 1,495 URLs sharing 2026-06-11T06:17:48+08:00: that is the commit that first added the curated-resources seed copy (Update built resources #795). The deployed files come from the daily data artifact, but Hugo still looks their paths up in Git, so every resource carried the seed's commit date.

Changes

• deploy.yaml and staging-aggregate.yaml: git config core.quotepath false after checkout (staging also gets fetch-depth: 0, matching production).
• config.toml: [frontmatter] lastmod = ["lastmod", ":git", "date", "publishDate"], so an explicit front-matter lastmod wins over Git.
• content/glossary/_create_glossaries.py and content/resources/resource.py now write a lastmod key (date only, UTC) and keep the previous value while an entry's content is unchanged. Only a real change to the entry moves the date. Shared logic in scripts/generated_lastmod.py. Bootstrap for entries that predate the key: the file's last commit date (glossary) or the sheet submission timestamp (curated resources).
• data-processing.yml: full-history checkout, and the curated generator compares against the build-resources branch, so a changed resource does not get re-dated daily until the next seed merge.
• scripts/check_sitemap_lastmod.py: reports undated URLs, any timestamp stamped on more than half the site, and (with --compare-source) pages whose sitemap date disagrees with their front matter or Git history. Runs warning-only in deploy on a 200-page sample.

Verification

Full local pipeline (both generators, Hugo build with quotepath off, checker on every mapped page):

Check	Result
URLs without lastmod	1 of 3,178 (the taxonomy root /publication_types/, which has no dated members)
Pages whose date disagrees with source	0 of 2,889
Generators re-run on unchanged data	byte-identical output

Things to be aware of

• One-cycle transient: the first deploy after merge still downloads the previous data artifact, so curated dates stay on the seed commit until the next daily data-processing run. Glossary dates come from Git until the next manual regenerate_glossary run is merged.
• Glossary bootstrap puts all ~1,300 terms on 2026-08-05, the commit that added en_title to every file. That is the last verifiable revision of each file, and it persists for terms that do not change afterwards.
• 297 curated resources have the placeholder submission timestamp 01/01/2020 in the sheet and will show 2020-01-01 as their date until their content changes. The previous value (the June 2026 seed commit) was equally uninformative, but this one is visible on the page.
• 28 hand-maintained pages (pedagogies, educators corner, summaries) set an explicit lastmod in front matter and now use it instead of their Git date. Their last commits were all housekeeping (link fixes, webp conversion), so the explicit values are the better ones.

🤖 Generated with Claude Code

[github-actions]

👍 All image files/references (if any) are in webp format, in line with our policy.

[github-actions]

✅ Spell Check Passed

No spelling issues found when checking 8 changed file(s)! 🎉

[LukasWallrich]

✅ Staging Deployment Status

This PR has been successfully deployed to staging as part of an aggregated deployment.

Deployed at: 2026-09-02 22:22:19 UTC
Staging URL: https://staging.forrt.org

The staging site shows the combined state of all compatible open PRs.

[richarddushime]

TO BE MONITORED LATER