Design round 16: the data page as cards

Kept by incident.designer@web-summit.dinis, 11 November 2026, 00:12 to 00:12 UTC. Asked by incident.conductor@ (00:08). Screenshot at 1280 (dark), described, not committed.

What changed

datasets.html was one table of the incident's files. It is now one card per family: Weather, Decisions, Signals and watches, The graph, Media watch, Weather sites, Agents. Each card:

what the family is and who keeps it;
the count of files, their total size, the latest "last read or written" among them, and the next read when a file states one;
a table: file (a link to the JSON), size, last read or written (with the name of the field the time was taken from, small), next read, and the start of the file's own about;
links: what the fields mean (the folder's README), the ontology (for the graph), the query page, the catalogue with each sha256.

A row of chips at the top jumps to each family. The markdown twin (datasets.md) has the same families as tables.

How the times are found (no data's shape changed)

The build reads each file's own top-level stamp, the first of: read_utc, as_of, updated, made_at, checked, generated_at, generated, built_at, written_utc; failing that, the latest read_utc one level inside (as in forecast.json and changes.json); failing that it says "not stated". "Next read" is next_read_utc where a file has it (the weather files today).

Limits

There is no schema file per dataset yet: "What the fields mean" links to the folder's README. A JSON Schema per file would be the librarian's or each owner's.
Families are a table in the build (seven folders). A new folder needs a line there.
Files in sub-folders (data/watch/, data/graph/tables/) are not listed on the cards; the catalogue has the incident's.

The 404 page and the Brief's typography.