Sources & privacy

Where the data comes from, what licence it carries, and what this site records about you. Everything below was checked against the running service rather than written from intent; where something is undecided it says so instead of implying a policy.

Where the data comes from

Geocoding — address lookups are served by a self-hosted Nominatim instance over an OpenStreetMap extract. That data is © OpenStreetMap contributors, licensed under the Open Database Licence 1.0 (ODbL). If you redistribute geocoding results or anything derived from them, the ODbL's attribution and share-alike terms travel with them. Nothing is proxied to OpenStreetMap's own servers, so their usage policy is not in play for your requests.

Census and federal data — the tract, county, state and PUMA tables are built from U.S. Census Bureau products (the American Community Survey, its Public Use Microdata Sample, and TIGER/Line geography) and from other federal sources joined alongside them: the FEMA National Risk Index, the Census Bureau's Community Resilience Estimates, USDA ERS Rural-Urban Continuum Codes, and NCES school finance surveys. Works of the U.S. federal government are in the public domain. Attribution is still the courteous thing to do, and describe_coverage on the MCP endpoint names the source and vintage of every column.

Which data, exactly — GET /health returns a data_version identifying the table being served, derived from the published artefact itself. Quote it if you need to say which data produced a number.

What this site records

Requests. Every request to the API is logged with its method, path, status, how long it took, and the calling IP address. The hosting platform additionally logs the full URL, including the query string, together with the IP and the browser's user-agent. This matters for the demo in particular: an address you type into the demo is written to a log, next to the IP that sent it. Those logs are kept for 30 days and then deleted automatically.

Accounts. If you sign up, we store a username, an email address, a hash of your password — never the password — the time you registered, and the display name and organisation if you provide them.

API usage. Calls to the MCP endpoint are metered. Each one records which account made it, which tool was called, the geography level, how many values were returned, how long it took and whether it succeeded. The arguments themselves are stored only as a SHA-256 digest, not as text. The geographies and variable names requested are recorded, so which places an account looked up is retained. There is no retention policy on this yet — deciding one is open work, and until it is decided those rows are kept indefinitely.

No tracking. There are no analytics, no tag manager, no advertising or social pixels, and the application sets no cookies.

One third party. Every page loads its typefaces from Google Fonts — fonts.googleapis.com for the stylesheet and fonts.gstatic.com for the files. That means your IP address and user-agent reach Google on every page view, under their privacy policy rather than this one. Self-hosting the fonts would remove that call; it has not been done yet.

What this page is not

This is a description of what the service does today, not a contract. It is an early prototype run by two people, there are no terms of service yet, and nothing here should be read as a commitment about a future version. If you need one before using this for something that matters, ask — the project is on GitHub.