We tuned a compact supply-chain forecaster from GPT-OSS-120B to predict disruption shocks from public news and trade-index context. The trained model beat GPT-5 across accuracy, calibration, and high-risk precision on the held-out evaluation set.
What we did
- Generated forecasting questions from supply-disruption index history plus prediction-time news context.
- Resolved each question from the next-month disruption outcome instead of hand labels.
- Tuned from GPT-OSS-120B with Foresight Learning and benchmarked against GPT-5 and the base model.
Read more
Primary artifacts for this case study.
Blog post blog.lightningrod.ai
From Metric to Prediction: Forecasting Supply Chain Disruptions with Foresight Learning
End-to-end walkthrough of training a supply chain disruption forecaster
Paper arxiv.org
Forecasting Supply Chain Disruptions with Foresight Learning
Method paper on outcome-based training for supply-chain forecasting
Dataset huggingface.co
Supply-chain predictions
Timestamped supply-chain questions with resolved outcomes