Halliday & Matthiessen (1999: 504-5):
The point to be brought out here is that there is no unique 'correct description' of a semiotic system, or of any phenomena that have to do with meaning. Stratal relations are not relations of cause & effect, and a stratified system cannot be reduced to constructions of content-expression pairs.
This is not to say, of course, when a choice is made among a set of alternative descriptions for representing features in grammar that are inherently indeterminate, like the types of process in a transitivity system, that the choice is insignificant, or merely random. On the contrary: it resonates throughout the grammatics as a whole (or should do, if the description has any claim to be comprehensive). Again there is the analogy between the grammatics and the grammar: just as no region of the grammar is isolated from the rest, so every descriptive statement has consequences throughout the description.