Halliday & Matthiessen (1999: 578-9):
In our analysis (unlike that of the mainstream grammatical tradition), the projected clause is not a constituent part of the mental or verbal clause by which it is projected. There are numerous reasons for this; some of them are grammatical — for example, it cannot be the focus of theme–predication … it cannot be the Subject of a passive mental clause … it is presumed by the substitute so, which is also used to presume conditional clauses in clause complexes … But these, in turn, reflect the semantic nature of projection: this is a relationship between two figures, not a device whereby one becomes a participant inside another.