Is a binary attributable/non-attributable label suffici...

While some benchmarks report a binary macro-F1 metric, this does not imply that binary judgment is the optimal design for a production pipeline. The FEVER dataset independently arrived at a three-class structure, confirming that a finer split is the natural granularity for fact-checking. Relying solely on a binary metric from a paper can lead to poor pipeline design that fails to handle the nuances of attribution.