Martin (1992: 433):
All this means that it is in principle possible to take a text and re-arrange it in any of the following ways without affecting cohesive harmony:
(i) Change the order of any two elements of the generic structure.
(ii) Weave the strings and chains realised as Subject or Theme into Residue or New.
(iii) Reverse the sequence of selected pairs of messages or moves.
Any of these disintegrating manœuvres would affect interlocutor's intuitions about the coherence of a text, which is simply to underscore the linguistic (as opposed to contextual), experiential (as opposed to interpersonal, textual and logical) and componential (as opposed to organic) bias in cohesive harmony analysis.
 None of this is true; see the previous post. All of these entail changes in the texture — the structural (thematic and informational) and non-structural (cohesion) resources of the textual metafunction, which in turn affects the cohesive harmony between the textual and experiential metafunctions.
 Here Martin presents his misunderstandings of cohesive harmony as a deficiency in the model itself; see the previous post and the next three points below.
 Cohesive harmony does not have a linguistic (as opposed to contextual) bias. Cohesive harmony is linguistic, not contextual. Cohesive harmony refers to the harmony of the resources of the textual metafunction with those of the other metafunctions in texts (language). Language realises context. They are two distinct levels of symbolic abstraction.
Of course, what Martin means here by 'contextual' is register and genre, both of which are (varieties of) language, not context.
 Cohesive harmony does not have a experiential (as opposed to interpersonal, textual and logical) bias; see the previous post. On the one hand, in 1985, it had not been extended to the interpersonal and logical metafunctions, and on the other, cohesive harmony takes the textual metafunction as its point of departure, since it is the textual metafunction that creates the texture with which the 'outputs' of other metafunction do or do not harmonise.
 Cohesive harmony does not have a componential (as opposed to organic) bias; see the previous post. Cohesive harmony does include the organic relations of non-structural cohesion but as Hasan makes clear (Hasan 1985: 85), she doesn't discuss these, or textual structures, for reasons of time and space.
This extract from Martin (1992) is intended as the second part of a critique of Hasan's cohesive harmony, as a prelude to introducing his own version. As can be seen from the clarifications above, it is merely a misunderstanding and misrepresentation of Hasan (1985).