How the hub works
The hub holds no research of its own. It reads a small, validated file from each site every night and shows what the file says, with its sources.
The stats contract
Each site publishes two files at fixed paths:
- A stats file: the site's name, tagline and scope; one headline fact and up to five more, each with text, value, unit, a named source (name, link, date) and the date it was pulled; record and source counts; the date of the site's last run; and the address of its map layer.
- A map layer: the site's findings as points, each with the finding's title, link, level, category and date. A finding known only to a U.S. state is placed at the state's centroid and labelled as placed.
A file is rejected, and the site's previous data kept, if any field is missing or malformed, if a fact lacks a source, link or date, if a date is in the future, or if the file names a company or carries a credit line.
Nightly refresh
A scheduled job runs twice a day (03:17 and 19:17 UTC). For each site in the registry it fetches the stats file, validates it, then fetches and validates the map layer. The result is recorded as a dated snapshot and deployed. The snapshot's time is shown on the home page. Every change to the figures is therefore a dated, kept record.
Stale data
If a site does not answer, or answers with a file that fails validation, the hub keeps that site's last good figures and marks the tile stale since the date of the last good fetch. Figures are never blanked and never guessed. After three days stale, the job raises an alert so a person looks.
Derived figures
Until a site publishes its own stats file, the hub derives one from the site's public data feed: counts of records and cited sources, and a map layer from record coordinates. Such a tile says derived by the hub and names the feed as its source. The site's own file takes over the moment it exists.
The map
Equal Earth projection over Natural Earth 1:110m land (public domain). One layer per site, toggled independently; a point's size on an aggregated layer reflects the count it stands for. State centroids come from U.S. Census Bureau cartographic boundaries via the us-atlas project. The map shows location only; it does not rank places.
Data access
The hub's figures, the snapshot behind them and the map layers are not published as open files. They are available on request, in JSON, to researchers, journalists, agencies and others with a stated purpose. Use the request form; each request is reviewed by a person.
What the hub never does
- Type a number in by hand. Every figure comes from a site's file or feed, through the validator.
- Rank sites, places or subjects against each other.
- Edit a site's figures. A correction is the site's to publish; the hub picks it up on the next refresh.
- Run a script from any third party other than the two pinned libraries on the map page.
- Publish its data or code openly. Both are available on request.