List your company

Metric Correlation with Human Judgment on Human-annotated Atomic Fact Attribution (test)

97CoverageMURGAT-SCORE
35.6451.5767.583.43Feb 12, 2026

Evaluation results

MethodLinks
MURGAT-SCOREBackbone (Decomposition)=Gemini-3-Flash, Backbone (Verifiability)=Gemini-3-Pro, Backbone (Entailment)=Gemini-2.5-Flash2026.029765590.86
DisentangledGranularity=sentence-level, Backbone=Gemini-3-Pro2026.027654500.58
DisentangledGranularity=response-level, Backbone=Gemini-3-Pro2026.025832490.45
HolisticGranularity=response-level, Backbone=Gemini-3-Pro, Score Range=1-52026.023839430.35