A separate strong model scores candidate outputs against a rubric, routing to approve / rework / reject.
Properties
family
Evaluation & Ranking
maturity
established
diagram
judge
whyMultiAgent
A judge distinct from the producer gives independent, rubric-grounded evaluation rather than a model grading its own work, strong judges match human preference ~80% of the time.
