A collaboration between Bridgewater’s AIA Labs and Thinking Machines Lab (founded by former OpenAI CTO Mira Murati) transformed proprietary expert judgment into high-quality training data, creating a custom fine-tuned model that achieves 84.7% average accuracy. This model delivers 29.8% fewer errors than the best frontier models while slashing inference costs by 13.8x. The work focused on six practical financial tasks drawn from investors’ daily workflows, including classifying whether a financial article is relevant to a C-suite macro investor and determining if a central bank document signals future interest rate moves. Frontier models like GPT, Claude, and Gemini had hovered around 46–50% accuracy on these tasks, with expert-crafted prompts only pushing them into the mid-to-high 70s. Bridgewater’s solution involved an iterative cleaning process starting with vendor labels, training an initial model, and routing disputed examples back to expert investors for re-evaluation. This “contentious rounds” approach filtered out errors and injected genuine expert patterns into the data, encoding subtle decision-making heuristics that experts intuitively apply but rarely spell out.