If the actual implementation use $N and is working, where is the complexity, and why would it introduces more bugs ?
It's hard to say, but judging by the use of $N, it's likely that mIRC uses some internal "tokenize" function on the resulting data, leading to $1- being filled-- if so, mIRC is using this tokenized data internally to return the data for the external \N API. That means that mIRC would have to change how it behaves internally, which clearly can lead to regressions that could even end up affecting $1- more globally, because one of these methods of refactoring the could would mean changing how the internal tokenize function itself works.
