AI-Human Comparisons: Direct experiments here mean lawyer-versus-system forecasting on US case outcomes, with attention to issue framing, procedural posture, and how much factual context each participant receives. CaseCruncher Alpha Analysis: The focus includes the Cambridge lab model’s architecture, feature representation, and recorded performance patterns, especially where legal reasoning was compressed into inputs a prediction system could actually use.
Legal AI prediction benchmarks are useful only when they respect the grain of litigation: uneven records, shifting doctrine, equitable judgment, and jurisdiction-specific procedure. A high score on a narrow dataset does not settle whether a model can support responsible legal decision-making in practice.