generate_images.py runs daily (.github/workflows/main.yml) and owns every number on the site. It writes four things from one API fetch:

Do not hand-edit those regions; the next run overwrites them. Everything else in index.html and llms.txt is hand-maintained.

The high-score table is generated from projects.json, which holds the editorial part: which projects appear, their label, link, initials and live-star. Score is stars * 1000 + forks * 100 and ordering follows it, so ranks track reality instead of drifting. Entries whose repo is absent from the stats fetch (forks and contributed repos are excluded by the query) fail the run loudly; set repo to null and give explicit stars/forks for those. Removing a data-stat attribute or a marker comment fails the run loudly rather than silently skipping the update.

A GitHub API call that cannot be answered now aborts the run instead of publishing zeros, so a failed workflow means "check the token or the API", not "the site has no stars". Transient rate limits and 5xx responses are still retried before giving up.

Run it locally with:

GITHUB_TOKEN="$(gh auth token)" GITHUB_ACTOR=boredland \
  EXCLUDED=boredland/boredland EXCLUDE_CONTRIBUTED_REPOS=false \
  uv run --with aiohttp python generate_images.py

Run the tests with:

uv run --with pytest --with aiohttp python -m pytest -q