I think another issue adding to the confusion in relation to theoretical vs descriptive categories in SFL is the rather loose use of notational conventions for descriptive categories. While (type of) processes (material, relational, etc.) refer to experiencial classes of clause (and they are thus expected to be written in lowercase), labels such as Actor, Process, Phenomenon (or more generic Participant, Circumstance) are elements in functional configurations of structure (and they should be written with initial uppercase). Too often in English descriptions and SFL literature such conventions are overriden, causing serious misundertandings. […]Of course, the discussion on theoretical categories such a as class, rank, function and structure leads us in turn to the question of whether the theoretical notion of rank-scale and its general bias towards constituency relations (at least in most SFL English descriptions) is indeed productive when looking at languages other than English. There is also very little discussion in SFL on this issue, although Fawcett 2000a, 2000b, 2000c does challenge the idea of a rank scale in English accounts (from the point of view of the Cardiff model), and Martin 1996 deconstructs the constituency bias in relation to the theoretical notion of structure (inspired by Halliday, 1979).
 The notational convention in SFL is to use lower case for the terms 'participant' and 'circumstance', since these are classes to which elements of function structure (Actor, Location etc.) belong.
 The rank scale isn't "biased" towards constituency; it is a theoretical means of construing language in terms of constituency — but one that goes beyond mere constituency, since it builds the function-form relation into the hierarchy, with function structures of a higher rank realised by syntagms at the lower rank. That is, the rank scale embodies not just composition, a type of extension, but also elaboration (and symbolic identity) through the realisation (intensive identifying) relation between ranks.
As Halliday & Matthiessen (1999: 146) make clear:
… in systemic-functional work, elaborating interpretations tend to be taken further than in many other approaches: this means emphasising realisation, delicacy and identities across metafunctions to supplement the traditional emphasis on constituency and composition.
The theoretical utility of the rank scale includes not only modelling the compositional relation from clause to morpheme, but also, for example, distinguishing embedding from taxis, and, most importantly, the unpacking of grammatical metaphor.