"Ethically sourced" is
a claim, not a vibe.
Most vendors using that phrase mean "we didn't get sued yet." Here it means four things you can check: every source named, every licence permits redistribution, no personal data, and nothing taken from organisations that publish under obligation rather than choice.
This page is the standard we hold ourselves to. If a pack ever falls short of it, that's a bug and we'd like to hear about it.
§01 — What we include
Two tiers, different obligations.
Tier A — Public domain
US federal government works, which carry no copyright. Redistributable without restriction.
- SEC EDGAR corporate filings
- US federal court opinions
- US patent grants and applications
- NOAA weather and climate observations
- Federal legislative and regulatory text
Tier B — Openly licensed, share-alike
Redistribution is permitted with attribution and licence inheritance. Those terms travel with the pack and bind you too — the licence text is in the archive.
- Stack Exchange data dumps (CC BY-SA)
- OpenStreetMap extracts (ODbL)
- PubMed Central Open Access subset (per-article CC licences, recorded individually)
- EUR-Lex EU legal texts (EU reuse policy, attribution required)
§02 — What we exclude
The exclusions are the interesting part.
Anyone can list what they included. The list below is what we left on the table, and why.
Anything licensed non-commercial
You are buying this to build something commercial. A CC BY-NC corpus would make that a licence violation the moment you shipped. We would rather not sell you the pack than sell you a lawsuit.
Charity, nonprofit, and NGO data
Form 990s and charity-commission filings are public, and turning a small organisation's mandatory disclosures into a product we profit from is not a thing we want to have done. Legality isn't the test we're applying.
Personal data of any kind
No voter files, no résumé scrapes, no health records, no breach dumps, no social profiles. Where a public record incidentally contains a name — a court opinion, a patent inventor — the pack is exactly that public record. Nothing enriched, nothing cross-linked, no entity resolution across sources.
Scraped forums and sites without an explicit data licence
"It was on the open web" is not a licence. If the operator hasn't published terms that permit redistribution, it isn't in a pack.
Novelty corpora
Public-domain fiction, sports archives, and similar are cheap to assemble and genuinely useful to almost nobody building a product. We only compile corpora that do a job someone is already paying to have done badly.
Anything we can't verify
If we cannot name the upstream source, produce its licence, and record when we fetched it, the record does not go in — even when we're fairly sure it would be fine.
§03 — What ships with every pack
PROVENANCE.json — every upstream source, its URL, its licence identifier, the date it was retrieved, and the number of records taken from it.
LICENSE.md — the licence terms you inherit, in plain language plus the full legal text. For Tier B packs this includes the attribution string you must carry and the share-alike obligation on anything you redistribute.
MANIFEST.json — file list, record counts, schema, and a SHA-256 for each shard so you can verify the archive arrived intact.
On energy. The packs are compiled and hosted on renewable-powered infrastructure. We are not claiming carbon neutrality — we haven't bought offsets, and saying otherwise would be marketing.
§ Reservation
Read the provenance before you read the price.
Every pack page lists its full source table publicly, before purchase. If the provenance doesn't satisfy you, don't buy it — and tell us what would.