Some queries for those of you introducing the term ‘subjunctive’.
Are you referring to a word class, a group class or a clause class?
If a word class, which group class is it realising; if a group class, which clause class is it realising; if a clause class, are there relevant clause complex relations to consider?
In each case, what is the relevant valeur, specified in terms of the entry condition to the relevant system and agnate features? And what is the realisation of the feature ‘subjunctive’?
Then, reasoning from above, what is the discourse semantic system that the ‘subjunctive’ class is realising? Are we talking about appraisal engagement [heterogloss: expand: entertain] options? And/or are we talking about connexion [cause: contingency: condition] relations?
 This misunderstands the architecture of SFL theory. In accordance with the principles of SFL theory, the term 'subjunctive' does not refer to a class of word (noun, verb etc.), nor to a class of group (nominal, verbal etc.), nor to a class of clause (adverbial etc.). The term 'subjunctive' is a potential feature in systems whose entry conditions are the rank of word, group or clause. Martin's focus here on classes of form, rather than functional features, takes a formal perspective, not a functional perspective.
 This misunderstands the architecture of SFL theory. The relation between word, group and clause is not realisation, but composition: a clause is composed of groups/phrases, a group/phrase is composed of words. With regard to the rank scale, realisation is the relation between function and form, as where an element of the function structure at clause rank, such as Process, is realised by a unit (simplex or complex) of the rank below, such as verbal group.
 According to Halliday & Matthiessen (2014: 143n), in Modern English, the subjunctive tends to be restricted to the dependent clause of a hypotactic nexus of condition or projection:
Many languages also have an interpersonal system of the verb(al group) that has been referred to as ‘mood’: it involves interpersonal contrasts such as indicative/subjunctive, indicative/subjunctive/optative. To distinguish these verbal contrasts from the clausal system of MOOD, we can refer to them as contrasts in mode. The subjunctive mode tends to be restricted to the environment of bound clauses — in particular, reported clauses and conditional clauses having the sense of irrealis. In Modern English, the subjunctive mode of the verb is marginal, although there is some dialectal variation.
 On the basis of Halliday & Matthiessen's characterisation, the relevant system is an interpersonal system termed MODE with the features indicative vs subjunctive. The entry condition includes 'verb' at the rank of word, and the feature 'subjunctive' is realised by a reduced set of forms, relative to the indicative.
 On the basis of Halliday & Matthiessen's characterisation, the grammatical feature 'subjunctive' realises the semantic feature 'irrealis'.
 The stratum of discourse semantics is Martin's invention only. As demonstrated in great detail here, it is theorised on fatal misunderstandings of SFL theory, with the result that it is inconsistent with SFL theory — as well as being inconsistent with itself.
 For some of the misunderstandings of the feature 'heterogloss' in the appraisal system of ENGAGEMENT in Working With Discourse (Martin & Rose 2007), see the explanatory critiques here.
 By 'connexion', Martin means his logical discourse semantic system of CONJUNCTION, which is his rebranding of Halliday's conjunctive cohesion, misunderstood, relocated from grammar to discourse semantics, and applied inconsistently to both logical structures and non-structural textual relations alike, as demonstrated in great detail here.