Data ethics & standards

1. Zero personal data

Every dataset records the institution, the law, or the designation — never private individuals. No personal data is published. Where a government source file names private people, the pipeline is designed so those names cannot reach the output: source columns are read against per-file whitelists that fail the build if the schema drifts, free-text and address fields are never ingested, person-name detectors abort regeneration on a match, and raw source files are never republished.

Where a register is inherently about individual holdings — foreign-held farmland, university donors — we publish aggregates only, and the only names that survive are governments, institutions, and entities named in the federal record itself.

2. The evidence ladder

A named entity requires a documented public-record fact and a source. We name a company or institution only where a government record or the entity's own filing does: a federal award, a regulator's designation, a court record, or the entity's own SEC filing. Every attribution carries its evidence tier and a link to the primary source.

Where the record stops, the dataset says so instead of guessing. The Detention Ledger is the exemplar: a facility's operator is named only where a federal award or the operator's own SEC filing attributes it — everywhere else the record reads “no private operator identified in federal records,” which is itself a finding. Nothing is asserted above its evidence tier.

3. Records, not allegations

These datasets restate official actions by named authorities, each with a primary-source link. We do not ourselves accuse any company or institution of wrongdoing; a designation, listing, or rating carries only the specific legal meaning its issuing authority gives it, and inclusion in a dataset does not imply wrongdoing. Neutral framing is part of the method: where a fact is contested, we attribute both sides to their sources.

4. Corrections & right of reply

Any entity named in a dataset may dispute an entry. Contact us with the entry and the primary record you believe contradicts it; we review disputes against the underlying government sources under the published corrections policy — acknowledged within 48 hours, resolved within 72 in most cases. Because every dataset is regenerated from its sources by script, a correction propagates to the published JSON, the page, and the machine manifest in the next build.