Estimating the Correlation Between Variables Under Destructive Testing, or How to Break the Same Board Twice