In our recent paper, we introduce AsymmetryZero, a framework for operationalizing human expert preferences as semantic evals.