From Entropy to Calibrated Uncertainty: Training Language Models to Reason About Uncertainty | Signal Canvas | ScienceToStartup