Built to
increase your trust in AI, not to feed doubt. It puts a
reliability quotient on any model's answer, so you can rely on the strong ones with confidence and verify only the few that need it. A high score is earned trust; a low score is a heads-up, not a verdict. It measures the reliability of the answer, not its truth. You decide.
How it works.
It measures agreement, not truth. A model confidently wrong the same way every time will still agree with itself, so consensus alone is not proof; surviving contradictory evidence and tracing to a source (not to an authority's stamp) is the next axis. BUILT: the corroboration meter, the fiction frame, the abstain, alternates on low corroboration. COMING: the contradiction / source axis.
Methods and limits.