The Evidence Register defines the labels used to evaluate claims across ChatGPT Disaster. It is designed to make source strength visible instead of forcing readers to guess whether a claim comes from a filing, a study, a news report, a company statement, or a user account.
Evidence Labels
- Primary source: court filings, regulatory records, official company statements, status pages, public datasets, transcripts, and original documents.
- Peer-reviewed research: published academic or clinical research, including journal articles and conference papers where methodology is available.
- Institutional report: reports from universities, nonprofits, government agencies, safety organizations, or standards bodies.
- Mainstream reporting: named reporting from established newsrooms, trade publications, or specialist outlets with editorial accountability.
- Company statement: statements, blog posts, release notes, policy updates, status pages, or public comments from AI companies or their executives.
- User-submitted account: direct submission, forum post, Reddit comment, social post, email, or interview account. These are treated as testimony unless corroborated.
- Editorial analysis: conclusions, rankings, comparisons, or interpretations drawn from multiple sources.
High-Stakes Claim Standard
Claims involving death, self-harm, psychiatric harm, medical advice, legal sanctions, financial loss, minors, private individuals, or criminal allegations should be tied to the strongest available source and updated when the public record changes.
What This Does Not Mean
A label is not a guarantee that a claim is final or undisputed. Lawsuits contain allegations. User reports contain personal accounts. Company statements can be incomplete. The label tells readers what kind of evidence is being used.