How We Know What We Know

This page is the plain-language, public distillation of the underlying policy documents. The full versions govern in case of any difference: research methodology, editorial policy, fact-checking policy, source policy, and AI usage policy.

How sources are ranked

Original sources are preferred over articles describing them. In order of preference: the Constitution and constitutional materials; statutes and regulations; court opinions; official government datasets; the Government Accountability Office; the Congressional Budget Office; the Congressional Research Service; inspectors general; federal and state agencies; legislative records; peer-reviewed research; established nonpartisan research institutions; reputable journalism; advocacy organizations representing competing positions; and public commentary or social media, used only as evidence of public reaction.

How facts are verified

Every material factual claim is tracked as its own record with a verification status: verified, verified_with_qualification, disputed, outdated, pending_review, unsupported, or rejected. Outdated figures are never deleted — they are preserved with their applicable date, and current public text is updated to the latest verified figure.

How political balance is approached

Balance does not mean pretending all claims are equally supported. A well-evidenced claim is presented as such even when it is politically inconvenient to one side — and that is stated openly, not hidden behind false equivalence. For genuinely contested claims, credible sources representing major competing interpretations are presented side by side.

How reforms are evaluated — the Responsible America Test

Every reform must have real, specific answers to all 28 questions below before it can be marked approved. This list is also the machine-readable responsible_america_test array in the policy taxonomy.

    How automation is used

    Automated research may surface candidate updates — new legislation, new court decisions, new agency data — into a review queue. It never publishes a substantive conclusion to this site without human review. AI-authored draft content is always created with publication_status: draft and fact_check_status: pending_review.

    How disagreement and correction are handled

    Genuinely contested claims are marked disputed and presented with competing sources rather than resolved by editorial fiat. Errors are corrected transparently: every reform's revision history (visible on its page) logs what changed, when, and why — nothing is silently edited.

    How to report an error

    See the correction-request path described in the citizen action policy. Viewpoint disagreement alone is never grounds for removing content; a specific, checkable factual error is.