Search papers, labs, and topics across Lattice.
The paper introduces "Foresight Learning," an end-to-end framework for training LLMs to forecast supply chain disruptions using realized outcomes as supervision. This approach addresses the challenge of reasoning about infrequent, high-impact events from noisy data, where general-purpose models often falter. The resulting models outperform GPT-5 and other baselines in accuracy, calibration, and precision, demonstrating improved probabilistic reasoning.
Forget prompt engineering – training LLMs directly on disruption outcomes unlocks surprisingly accurate and calibrated supply chain forecasts, even beating GPT-5.
Anticipating supply chain disruptions before they materialize is a core challenge for firms and policymakers alike. A key difficulty is learning to reason reliably about infrequent, high-impact events from noisy and unstructured inputs - a setting where general-purpose models struggle without task-specific adaptation. We introduce an end-to-end framework that trains LLMs to produce calibrated probabilistic forecasts using realized disruption outcomes as supervision. The resulting model substantially outperforms strong baselines - including GPT-5 - on accuracy, calibration, and precision. We also show that training induces more structured and reliable probabilistic reasoning without explicit prompting. These results suggest a general pathway for training domain-specific forecasting models that produce decision-ready signals. To support transparency we open-source the evaluation dataset used in this study. Dataset: https://huggingface.co/datasets/LightningRodLabs/supply-chain-predictions
