Distinguishing Characters

[julian]

Yes, such a cipher might work: we know the VMS "word" lengths are binomial distributed, so that fits with them being numbers. A couple of properties would need to be reproduced: a) removing the first glyph/character/symbol from the word often makes another valid word (this works for your XIV45 example), and b) (more difficult) the frequent occurrences of word repetitions. For b) it's hard to explain those if there is a one to one mapping between the plaintext and the ciphered form, right?