Base120 · Recursion
Calibration Loops
Repeatedly check predictions against outcomes to improve forecasting accuracy
When to use
"Use when making repeated predictions. Track predictions against outcomes, measure calibration (are you right as often as your confidence suggests), and adjust. Improves forecasting accuracy over time."
Example
"A team estimates project timelines. They track — predicted 2 weeks, actual 3 weeks; predicted 1 week, actual 1.5 weeks; predicted 4 weeks, actual 6 weeks. The pattern (actual = 1.5x predicted) is a calibration error. Future estimates multiply by 1.5."
Common misuse
"Tracking predictions without tracking confidence. A prediction of 2 weeks at 90% confidence and at 50% confidence are very different. Calibration requires both the prediction and the confidence level."