Methods that make model predictions understandable to humans, including feature attributions and counterfactuals, important for regulated DeFi and identity scoring.