As far as I can see the section on negative numbers was completely made up, or at least it is not sourced in the system's patent. The encoding was mostly defined as a way to transmit (and write down) binary data in nibbles (groups of 4 bits), it does not really concern itself with what that data encodes.
The one thing it does support which the wiki does not note, is a run-length encoding: if a syllable ends in R, then the segment until the next R is interpreted as a positive integer and a repetition count so HORBAR is 05 (5 repetitions of 0), and DIRHAHOR is 1516 (16 repetitions of a fully set nibble).
And of course I forgot about HN's shitty markup and didn't notice before the edition timer ran out.
HORBAR is repeat(0x0, count=5)
DIRHAHOR is repeat(0xF, count=16)
Also in script form this is written as a strange parenthetical. Strange because the repeated nibble gets an open paren "stuck" to it, like a tail. So repeated nibbles don't look much like their base form.
The one thing it does support which the wiki does not note, is a run-length encoding: if a syllable ends in R, then the segment until the next R is interpreted as a positive integer and a repetition count so HORBAR is 05 (5 repetitions of 0), and DIRHAHOR is 1516 (16 repetitions of a fully set nibble).