Calibration Loop

Origin. The structure appears wherever prediction matters. Shang oracle bones recorded not just the divination but later outcomes, comparing what was predicted to what occurred. Scientific method institutionalizes hypothesis-test-revise. Reference class forecasting uses past outcomes to correct future estimates. Weather forecasting, medical diagnosis, and any mature predictive discipline. The pattern persists because judgment that never checks itself drifts from reality.

Mechanism. Unaided judgment is poorly calibrated — people are overconfident, underconfident, or systematically biased in ways they do not notice. Calibration requires feedback: you must discover how your predictions compare to outcomes. The loop closes when outcomes inform future predictions. Over time, calibrated judgment learns its own reliability: when to trust itself, when to seek more information, when to defer. Without the loop, judgment operates open-loop and errors accumulate.

Procedure. Make the prediction explicit and recordable before the outcome is known. Specify what would count as the prediction being correct or incorrect. Wait for the outcome. Compare prediction to outcome. Analyze discrepancies: was the error random or systematic? If systematic, identify the source of bias. Adjust the prediction process. Repeat. Track calibration over time: are predictions improving? The discipline is in recording before, comparing after, and actually adjusting.

Applies to. Any domain involving prediction: forecasting, diagnosis, estimation, risk assessment. Personal decision-making under uncertainty. Institutional judgment where accountability matters.

Limitations. Requires outcomes that can be observed and compared — some predictions concern events too distant or too ambiguous to check. Feedback must be timely enough to inform future predictions. The method improves calibration but does not guarantee accuracy; you can be well-calibrated and still wrong. Recording predictions requires discipline; most people skip this step. Some domains have slow feedback loops that make calibration difficult.