Engineering your own voynich

[oshfdk]

I have a non-ambiguous 3-way partition of the words of byatan's cipher in 26 different strings for each part, where the second and third parts can be empty, that works as a fully deterministic and reversible cipher. The requirement for deterministic reversibility is, for the strings ("", "aei" and "z") of ciphertext that can appear in both part 2 an part 3, to represent the same cleartext in both positions 2 and 3.

I like it! Especially convincing given the last token is the only one of 2 characters long, just because the algorithm ran out of plaintext.

It's suggestive that there are 26 different strings for each of three positions, is this the most obvious split, or were you specifically aiming for 26 per slot? This can't be a static map over the English alphabet, unless the text is about foxes jumping over lazy dogs. Maybe the character encodings are updated at each turn according to some algorithm?