Open Graph card, a footer link to the source, and the repro command above the fold

[ChelseaKR]

Three things that sit between a reader and a number, none of them about the measurement itself.

The link unfurls as a grey box

The pages already carried an Open Graph block — og:title, og:description, og:url, og:site_name, og:locale — with no image in it, and twitter:card set to summary. Every consumer of those tags falls back to a blank rectangle when og:image is absent, so the one channel these measurements actually travel through, somebody pasting the link, showed nothing about them.

Adds og:image (absolute; Open Graph consumers resolve it with no page context and drop a relative one) with og:image:type, :width, :height and :alt, and moves twitter:card to summary_large_image with a matching twitter:image and twitter:image:alt.

The card is 1200×630, drawn in the site's own light palette and its own serif, carrying the title, the one-line description and the three-state key the tables use. Nothing on it is a number, so it cannot go stale against a retrieval.

Where the card lives, and why not in site/

It is a committed source asset in assets/, copied into the output by build_site — not a static file dropped into site/.

make site-check diffs site/ against a fresh build from the acquired files. A file served out of site/ that no build writes would report as drift on every single run, and a check that always says the same thing stops being read. Copying it in the build also puts the card in build/site-offline and build/run-one/run-two, which is where the page checks and the determinism gate look at the pages that reference it — determinism now compares 6 files instead of 5.

The site never said where it was built

The footer credited CAL FIRE's licence and carried the unofficial disclaimer, and named the build nowhere. These pages publish inferences — docs/MARKERS.md exists precisely so a reader can check one — and a reader who wanted to had no route off the page to the field registry it came from. One footer line, rendered on all three pages.

The repro command was at line 129

Three hosted report pages are linked at line 5 of the README; uv sync and npm ci did not appear until line 129. Anyone wanting to verify the numbers rather than read them had to scroll past the whole methodology first.

make site-offline now sits next to the hosted links: it is the real pipeline over the committed fixtures, needs no network and no CAL FIRE download, and runs in about a minute. The block names the real-data path (make acquire, make site) and points at make site-check as the step that actually settles whether the served numbers are what this code produces today. The hosted reports still lead — this is four lines, not a relocation of the Build section.

Verification

site/ was rebuilt with make site from the acquired FRAP and DINS files rather than hand-edited. The diff is exactly the 9 head lines and 3 footer lines per page; no published count moved, which is its own small confirmation that the pipeline still reproduces the committed measurement.

• make verify — passes end to end: lockfile, lint, format, mypy --strict, 690 tests, pip-audit, html-validate, axe-core in jsdom and in Chromium, WCAG 2.2 reflow at 320×256, npm audit on both toolchains, determinism.
• make site-check — site/ is byte-identical to a fresh build from data/raw/.

Not fixed here

chelseakr.github.io is a shared origin serving several project sites from subpaths, so no project on it can own /robots.txt. Not this repo's to fix.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Pvm6FCdDXwtzA3juGxz6ba