Factuality
Does the claim hold against authoritative sources?
Machine Alignment
Eight dimensions of machine credibility. Each module produces its own record, so a composite score can always be decomposed back into the evidence that made it.
Does the claim hold against authoritative sources?
Is stated confidence proportionate to the evidence?
Can the statement be traced to an identifiable origin?
Which framings and groups are over- or under-represented?
Does the result survive rephrasing and adversarial input?
Do parallel statements agree across scopes?
How much of the relevant material was actually examined?
Does the score reproduce across runs and over time?