Data overview
The registry as files: datasets, formats, slices, tiers, license and how to cite.
In short
Everything public in the registry is published as files every night: seven datasets, each as gzipped CSV and gzipped JSON lines, listed with their row counts, sizes and sha256 in /data/manifest.json. Any dataset can also be cut to one industry or one upstream source on demand. Download from /data, with systatlas download, or over HTTP.
Datasets
Rows join on ids: components.organization_id, values.component_id, identifiers.organization_id, organization_links.from_organization_id, and organizations.industry_key to industries.key. Every field is on the data model page.
What is in the files, and what never is
- Organizations of the public registry: those with public parts or a verified claim.
- Parts, figures and links that are
publicorpublished, neverprivateorshared; never one the organization disputed, one taken down, or one rejected in review. A figure also needs its part to be public. - Sources that a public part or figure cites. A client's own documents and interview notes never leave, not even as a link on the row that cites them.
Formats
- CSV (
.csv.gz): UTF-8, a header row, RFC 4180 quoting. Lists join with;, JSON fields are JSON text. A text cell that starts with=,+,@or-(not a number) gets a leading'so a spreadsheet never runs it. - JSON lines (
.jsonl.gz): one object per line, values as they are. - Times are UTC ISO 8601 (
2026-11-03T03:00:00Z), datesYYYY-MM-DD. Rows are ordered by id, so two runs diff cleanly.
Slices
A slice is one dataset for an industry (the node and everything under it in the ontology, on an organization's primary or secondary placement) or for one source (the rows that source gave us, with its attribution). Slices stream from the database as they are downloaded, in the same formats and columns as the bulk files.
Who can download what
Every download goes through a signed link that lasts 5 minutes. Ask for one with your session or an API key; links are rate-limited per account (per address when signed out).
curl -s "https://systematlas.org/api/data/link?dataset=values&industry=naics_6212&format=csv" \
-H "Authorization: Bearer sa_…"
# { "data": { "url": "https://systematlas.org/data/dl?t=…&cap=10000&exp=…&sig=…",
# "name": "values-industry-naics_6212.csv.gz", "rows_cap": 10000, … } }The manifest
/data/manifest.json lists the files of the latest run with rows, bytes, sha256 and the time each was written, its access, and the rows each upstream source contributed with that source's license and attribution. The run before stays downloadable until the next night.
{
"version": 1,
"generated_at": "2026-11-03T03:00:00.000Z",
"license": { "name": "CC BY 4.0", "attribution": "System Atlas (systematlas.org)", … },
"upstreams": [{ "key": "gleif", "license": "CC0 1.0", "attribution": null, … }, …],
"files": [
{
"dataset": "organizations",
"format": "csv",
"name": "organizations.csv.gz",
"rows": 110412,
"bytes": 9811034,
"sha256": "4f1c…",
"generated_at": "2026-11-03T03:00:04.112Z",
"access": "open",
"license": "CC BY 4.0",
"upstreams": [],
"download": "/api/data/link?dataset=organizations&format=csv"
}
]
}License and attribution
The compilation (which organizations, how they are mapped, the parts and figures we found) is ours, under CC BY 4.0: use it for anything, name System Atlas and link systematlas.org.
Rows from an upstream whose license asks for attribution carry it in an attribution column, and name their source in upstream: keep both with the rows you reuse. CC0 and public-domain sources need none.
Quoted words in quote belong to the page they were read from; we keep one sentence at most, with its link.
How to cite
Name the run you used (its date is in the manifest):
System Atlas registry (2026-11-03). systematlas.org/data. Licensed CC BY 4.0.