Scorecard builds simulation environments and reward models that frontier AI labs and enterprises use to train and evaluate their agents. Same discipline that made self-driving cars work, applied to LLM-based agents.