I was hoping to see there a block with unregistered domain names. One could do it by, for example, training a Markov chain/neural network to spew out (sample)

I was hoping to see there a block with unregistered domain names. One could do it by, for example, training a Markov chain/neural network to spew out (sample) domains, and then see how many are currently registered, to compute the general XXXX.com space utilization fraction. This approach would be powered by a more sensible, IMHO, probability distribution than, say, taking all domain names with up to 30 {a-z0-9} characters with uniform probabilities.