That's a great essay, but the author is talking about random-walk situations, which are independent random variables. We know this, because that's how the automatic software testing tools are (currently) written.
Correct, I probably should have pointed out more context here.
The reason that random testing in this fashion would be a problem is only if the actual distribution of requirements does not well match the central limit theorem -- if for example user demand is very correlated. Because the central limit theorem is usually off, it is not acceptable to assume that testing independent random variables will reproduce adequately the risky situations testing should test.
