How TruthRate grades evidence
A grade says how strong the evidence is for a claim about an AI system. It does not say how serious the harm was or how bad the system is.
Verified
The event and the AI system's role are both documented by a primary source, such as a court or tribunal decision, a regulator or inquiry report, or a peer-reviewed study with data.
Example: Moffatt v. Air Canada: the tribunal decision reproduces the chatbot's answer, the conflicting web page and the fare calculation, and finds the airline liable.
Probable
Credible reporting or partial primary evidence supports the claim, but a key link, usually the system's exact role, rests on a secondary source or a single account.
Example: A news investigation quotes insiders and shows documents, but no regulator, court or study has confirmed what the system did.
Contested
Credible sources reach conflicting conclusions on whether the system caused the harm. Both sides are shown.
Example: Independent analyses of the same tool disagree because they define fairness differently, and no authority has settled the question.
Insufficient
The claim was made publicly, but the available evidence does not support it, or no primary source shows the system's role.
Example: The Apple Card allegation: a regulator analyzed nearly 400,000 applications and found no evidence of sex discrimination, so the claim is not supported even though it was widely repeated.
Retracted
A source the case relied on was withdrawn or corrected, or the claim was shown to be false. The case stays visible with the reason.
Example: A study cited in a case is withdrawn by its publisher, so the case is re-graded and keeps a note explaining why.
Every case shows its sources, labeled primary or secondary. A case cannot be published without a claim, evidence, a causal-strength statement and at least one source.
Grades change when evidence changes. A reviewer approves each grade before it is shown.