At a very basic level - it’s fine. No approach is intrinsically more ‘accurate’ in terms of PBR than the other.
Once the material has been compiled into a shader, it’s treated in the same way by the engine with regards to performance / rendering - so if you are happy with the look you can use whatever creation methods you want.
The issues you might run into (and why you might be reading that it’s good NOT to mix techniques) is in asset re-use across applications, and consistency amongst the artists working on your project. You might also miss out on some potential opportunities for automation, again because of your lack of consistency.
It should also be possible to convert Roughness/Metallic maps into Spec/Gloss equivalents (and vice-versa) so there shouldn’t be any reason to mix the two. It’s just confusing.
However, if it’s just you working on it, and it’s looking how you want it to, then go ahead. You aren’t breaking anything. 