Your question is Evaluation with Subjective Ground Truth. 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 do you handle model evaluation metrics when the ground truth is subjective or difficult to quantify?
Treat this as a practical evaluation-design problem rather than a request for one universal metric. Explain how you would define success, construct or validate labels, select complementary metrics, quantify evaluator disagreement, and monitor reliability over time. Address how your approach changes when no single objective ground-truth label exists.