Halliday & Matthiessen (2014: 506-7):
Metaphenomena – projections – can be associated only with certain types of process, essentially saying and sensing, plus in certain circumstances being. Macrophenomena – expansions – can enter into material processes. Thus you can say [[ = crushing him like that]] broke his bones. But you cannot say it broke his bones that you crushed him like that, because finite that (‘indirect’) clauses can only be projections, not expansions. (You can on the other hand say it broke his heart that you crushed him like that, because heart-breaking, unlike bone-breaking, is a mental process.)
Complication arises because the names of metaphenomena, nouns such as belief and fact, can sometimes enter into material processes where the metaphenomena by themselves cannot. For example, although we cannot say it destroyed his life that the experiment had failed, we can say the knowledge that the experiment had failed destroyed his life – not the idea as such, but his knowledge of it, was the destroyer. We may also note abstract material processes used metaphorically to construe mental phenomena:The passage of time, romantic travellers’ tales – of which Marco Polo’s supply the classic example – and wishful thinking, all combined to build up the late medieval belief [[ that Prester John was a mighty, if probably schismatical Christian priest-king ]].
We might also say the fact that the experiment had failed destroyed his life; here fact stands for a state of affairs, rather than for a projected metaphenomenon as in its prototypical sense. In other words, although projections cannot participate in processes other than those of consciousness, the names of projections can, because they can be used to label events or states of affairs. Here we have reached the borderline between expansion and projection; the two come together under conditions of nominalisation, where there is metaphor in the grammar and many of the semantic distinctions expressed in the clause tend to be neutralised.