Claims with a real, disputed, or evolving answer — where the models can reasonably differ, or simply be wrong. Ranked by cross-model divergence — the questions where the machines disagree most are first.