[lazamar]

Indeed, but worth noting that LZ is a modelling scheme, whilst Huffman is a coding technique.

That is, LZ determines, dynamically as it goes, what are all the elements we want to encode and their probabilities. Then you need a coder, like Huffman, to actually encode it.

In the post I used a semi-static zero-order byte-based model. Which means I counted the byte occurrences first and just used that count for the probabilities throughout all of the encoding. Then I used Huffman codes to translate those probabilities into bits.

But I'm considering writing a follow-up changing this static model for an LZ77 one as I think that would be fun.