Comment on: Foresight by Lightning Rod

The OpenAI-compatible interface is the right call. It means teams can drop this into existing agent pipelines without touching their orchestration layer. We've hit the same problem with general LLMs hallucinating probabilities. They'll say '70% chance' with no calibration behind it. How do you handle calibration drift as events resolve? Is accuracy validated continuously against a live benchmark, or is it a periodic evaluation cycle?