Martin (1992: 371):
The main differences between nuclear relations and cohesive harmony analysis is that Hasan's approach is based on lexical items rather than message parts and built up around TRANSITIVITY relations rather than extending and enhancing ones. This is because nuclear relations obtain at the level of discourse semantics in the model being developed here, with message parts realised by one or more lexical items and the elaboration, extension, enhancement analysis stratified with respect to TRANSITIVITY and group rank experiential grammar.
 The main difference between Hasan's cohesive harmony analysis and Martin's nuclear relations analysis is that Hasan's model is consistent with SFL theory — modelling how the textual and experiential metafunctions harmonise — whereas Martin's model is neither consistent with SFL — being located within the experiential metafunction, confusing cohesion with lexis etc. — nor self-consistent. See previous posts for details.
 That is, Martin's development of the textual resource of lexical cohesion involves experiential units related logically by two of the three types of expansion.
 The difference between the two models is not simply a matter of stratal location and realisation relations. Martin takes a non-structural resource of the textual metafunction at the level of lexicogrammar and "develops" it as a structural resource of the experiential metafunction at the level of discourse semantics.
 That is, structural units, at the level of discourse semantics are realised by one or more non-structural (lexical) outputs at the level of lexicogrammar.
 That is, in this development of the SFL textual system of lexical cohesion, logical relations at the level of discourse semantics are "stratified with respect to" experiential systems at the level of lexicogrammatical stratum — with the realisation relation between them left unspecified.