"Because we cannot yet (1) characterize all the possible experimental designs along quantitative scales and (2) generate cost-of-error functions, comparisons must be made in specific contexts rather than by use of analytic optimizing."