How to cite us — and check us
Every figure names its source, labels how sure it is, and is served in machine formats you can fetch. Don't take the page's word — re-fetch the source.
Atlas Commons is built to be cited and checked, not believed. Every figure is a research object that names its source, labels how sure it is, and is served in machine formats you can fetch directly. You do not take the page's word — you re-fetch the named public source and verify. This page is how.
Agent-native formats
Every entity is projected to these formats from one definition, so they cannot disagree — fix the claim, fix every
format. Append the extension to any entity URL (e.g. a state page → …/states/fl.jsonld).
-
.jsonldapplication/ld+json — the canonical research object — schema.org-typed, with each claim's warrant, source, and the structured tables in hasPart. -
.jsonapplication/json — the same research object as plain JSON, for agents that don't want JSON-LD framing. -
.mdtext/markdown — the claims with their warrants and sources, plus the tables — for an LLM reading context. -
.tomlapplication/toml — the same, as TOML, with the structured tables as arrays of typed rows. -
.context.mdtext/markdown — a condensed, citation-first brief that points back to the full .jsonld.
Format coverage — honest rollout status
Live The flagship analytical pages and the per-state risk profiles carry all five formats, projected from one definition by the @f3/atlas-research-objects framework — so the page and every format are the same research object and can never drift.
Rolling out The long-tail entity routes (individual facilities, companies, dockets, counties) currently carry .jsonld and .md; the full five-format projection is being migrated across them route-family by route-family. Every route already carries at least the .jsonld research object and the .md.
How to cite a figure
Carry these fields so the figure stays verifiable away from the page:
- the claim @id — the stable identifier of the specific figure within the research object (the object's @id + a claim fragment).
- the warrant — its four-status label — RECORDED / MODELED / DERIVED / INTERPRETATION — so the reader knows what KIND of figure it is.
- the named source — the public authority and dataset the figure resolves to (e.g. FEMA National Risk Index), with its URL.
- the re-fetch — the source URL you (or your reader) can pull to reproduce or challenge the figure — Atlas Commons is the index, the authority is the source.
- the dataset version — the research object's version / the source release the figure was taken from, so a later change is detectable.
FEMA models Florida's expected annual loss from natural hazards at $8.79B/yr (MODELED; FEMA National Risk Index, hazards.fema.gov/nri/). — Atlas Commons, /property/states/fl.jsonld
The provenance lineage
Every figure traces this chain from a public source to the served format — and every stage is inspectable:
- 1
The public source
A named, authoritative public dataset (FEMA NRI, NOAA Storm Events, CMS, OpenFEMA, the federal courts, …). Atlas Commons never invents a figure; a language model is never the source.
- 2
The medallion gold
The source is ingested and aggregated into a content-hashed gold table with a recorded generation time — the reproducible transformation between the raw source and the published figure.
- 3
The research object
Each consequential figure becomes a four-status claim — what occurred, what's supported, what's adjudged, what's operative — carrying its source, its warrant, and a falsifier that says how it could be shown wrong.
- 4
The projection
One definition is projected to every machine format (and the human page). The formats are pure projections of the same object, so they cannot disagree — fix the claim, fix every format.
What is guaranteed — and what is not
Guaranteed
- Every consequential figure resolves to a named public source you can re-fetch — never to a language model and never to an unnamed origin.
- Every figure carries a visible four-status warrant; a modeled expectation is never dressed as a recorded fact.
- The page, the .jsonld, and the other machine formats are projections of one definition — they carry the same figures, by construction.
- Corrections are logged publicly; what was wrong, and the live corrected surface, are on the record.
Not guaranteed — read this
- Atlas Commons cites the source RELEASE it ingested (with a generation time and content hash) — it is not a live pass-through. The upstream source can change after that release; re-fetch the source for the current value.
- Atlas Commons does not vouch for the source's own accuracy. It faithfully indexes and labels what the authority published; if the authority is wrong, the citation lets you challenge the authority, not the index.
- Derived figures (rankings, ratios, per-capita rates) are the atlas's own transparent calculations over the recorded source — labeled DERIVED, with the method stated, not an authority's verdict.
For agents. This page is machine-readable at /provenance.jsonld
(and /provenance.md). The epistemic model in depth — the four-status warrant,
the falsifier, why you re-fetch the source — is at the research object.
It is part of the Public Knowledge Charter; what we know is wrong is on the corrections ledger.
Agentic University · Provenance v1 · CC0 · the research object · the charter · machine-readable at /provenance.jsonld