Martin (1992: 19):
Halliday and Hasan do recognise that "cohesion within the sentence need not be regarded essentially as distinct phenomena" (1976: 9), but argue that in the description of a text "it is the intersentence cohesion that is significant, because that represents the variable aspect of cohesion, distinguishing one text from another" (1976: 9). In a sense, the line they draw between cohesion and grammar is justified descriptively rather than theoretically.
 Here Martin misinterprets the line between 'cohesion within the sentence' and 'intersentence cohesion' as the line between cohesion and grammar. Both uses of cohesion are cohesion, the non-structural component of the textual metafunction, and cohesion is located on the lexicogrammatical stratum.
 The distinction between 'cohesion within the sentence' and 'intersentence cohesion' is thus not a descriptive justification of a distinction between grammar and cohesion. The reason Halliday and Hasan focus on 'intersentence cohesion' for the description of a text is that this is most relevant for comparing texts in terms of (as instances of) register. The quote from Halliday and Hasan (1976: 23) that Martin provides (op cit: 28) makes this clear:
The concept of COHESION can therefore be usefully supplemented by that of REGISTER, since the two together effectively define a TEXT.