CalCOFI Usage

Which CalCOFI apps, sites and services people actually use — published at https://calcofi.io/analytics/ and refreshed daily from Google Analytics. Each card on calcofi.io links here from its usage link, next to its status link into status.calcofi.io.

Privacy — what is collected

The site publishes aggregate counts only. Specifically:

• No IP address is published, and none is even available. Google Analytics resolves visitor geography on its own servers; the Data API returns country / region, never an address. The map is built from those.
• The db-viz-hex query log is the one place personal data exists — that Sheet carries ip and user_agent per row, because the app records them server-side. scripts/sheet.py drops those columns as its first action, uses the id columns only to count distinct visitors, and asserts on the way out that none of them reached a written file.
• Small cells are suppressed: a region with fewer than 10 active users is folded into “Other”, on top of Google's own thresholding.
• Free-text is capped: the "most-requested" table publishes the top 20 values only, so a one-off query cannot become a fingerprint.

Every number on the site is also downloadable as CSV under /analytics/data/, so any claim here can be checked.

How it works

GA4 (2 properties) ─┐
                    ├─ scripts/refresh.py ─→ static/data/**.csv ─→ scripts/build.py ─→ data/usage/*.json
db-viz-hex Sheet ───┘        (attribution)      (accumulated)                          content/products/*.md
                                                                                              │
                                                                              hugo ───────────┘──→ calcofi.io/analytics/

• data/registry.yml is the only file to hand-edit. It maps each product slug to a GA4 property and the rule that claims its rows.
• CSVs are accumulated, not re-queried: GA4's default retention is 2 months, so re-fetching the whole history each run would quietly lose the past. Each run re-reads a trailing 35-day window and upserts by date.
• Attribution is contentGroup first, page path second. contentGroup is a built-in Data API dimension already emitted by every CalCOFI app (calcofi4r::cc_ga_js) and by the static sites via snippets/gtag-site.html. The path fallback covers Quarto/pkgdown pages that emit no group of ours, plus each product's history from before it was tagged.
• Not app_name: it is an event parameter, so the Data API can only reach it as customEvent:app_name after someone registers a custom dimension in the GA4 UI, and it backfills nothing.

Two properties feed the site — one for the Shiny apps (G-VV117EV9ZT), one for the calcofi.io sites (G-0HVK8TDMCF) — and they cannot be de-duplicated against each other, so org-wide totals count a person who visits both once per property. The index page says so in a footnote.

Local development

pip install -r scripts/requirements.txt
python3 scripts/build.py     # regenerate pages + JSON from whatever CSVs exist
hugo server                  # http://localhost:1313/analytics/

build.py runs with no credentials and no data — the site builds and every /analytics/<slug>/ URL resolves, showing "no data yet". That is deliberate, so the links from calcofi.io are never broken while the data catches up.

To pull real data you need the Google setup in OPERATIONS.md, then:

GCP_SA_KEY="$(cat key.json)" USAGE_SHEET_ID=... python3 scripts/refresh.py --backfill

Why Hugo

This is the org's first Hugo site; its siblings (db-schema, db-query, CalCOFI.github.io) are Jekyll. The layouts here deliberately mirror theirs — same palette, same pre-paint theme script, same dual-logo header — so that if the others migrate, the port is mostly copy-paste.

Charts are inline SVG rendered by Hugo at build time; the only client-side code is assets/js/app.js, which adds hover tooltips (one delegated listener for the whole page), table sorting and the Leaflet map. There is no charting library.