So this is essentially short complete Markov chain with recovery? What is the secret to performance in this probabilistic compressor? Word size alignment and locality?

Is this based on new extended variant of Ross's theorem? (multivariate case)