Many self-* properties are variations on the same theme: resilience of a system to changes in itself or the conditions under which it operates. Quantifying resilience is difficult, however: there are no metrics of resilience that are readily comparable across systems, and the space of possible changes is typically prohibitively large. To address this problem, I propose a quantitative measure of graceful degradation that is independent of the units, scales, and number of system parameters. Although this metric is typically intractable to compute precisely, it can be approximated by perturbation surveys, and the quality of approximation is likely to be improved by a random perturbation approach based on recent advances in manifold learning.