Your question is Preference Benchmarks With Automated Scoring. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you use automated evaluation frameworks to score model outputs against human preference benchmarks?