In comment 58 to my Actually Observed thread ID opponent keith s shows ID opponent Learned Hand how the design inference works.

To use the coin-flipping example, every sequence of 500 fair coin flips is astronomically improbable, because there are 2^500 possible sequences and all have equally low probability. But obviously we don’t exclaim “Design!” after every 500 coin flips. The missing ingredient is the specification of the target T.

Suppose I specify that T is a sequence of 250 consecutive heads followed by 250 consecutive tails. If I then sit down and proceed to flip that exact sequence, you can be virtually certain that something fishy is going on. In other words, you can reject the chance hypothesis H that the coin is fair and that I am flipping it fairly.

Keiths then goes on to with some mistaken interpretations of Dembski’s work. But that’s all right. The important thing is that even one of our most inveterate opponents agrees with the basic thrust of the design inference. We are making progress.

Ironically, later in the post Learned Hand writes:

I think I’ll start a clock on any ID supporter actually testing whether CSI can detect design without knowing (or assuming) in advance whether the subject is designed.

Well, LH, he is far from an ID supporter, but will keiths example work?