Martin (1992: 261-2):
The text is repeated below to highlight the significance of regressive dependency and depth in RST analysis. Note for example that the second message, Tempting as it may be, is a satellite to [4:199c], which is in turn part of a satellite to [4:199g-i], which is in turn part of a satellite to [4:199n]. Before this clause could be generated in other words the text would have to be processed backwards from its end message to one of its deepest points. …
It would perhaps be unfair to suggest that RST analysis treats [4:199] as a text that unfolds backwards from its end instead of working forwards from is beginning. The picture is in fact more complex than this, involving both progressive and regressive dependencies. However, it is clear that Rhetorical Structure Theory is not at all concerned with the way in which text unfolds as a process, with what has been realised taken as a point of departure for what comes next. Somewhat ironically, given the similarity between the RST interpretation and clause complex analysis, rhetorical structures are very synoptic in orientation. They account for text as a finished, pre-conceived, edited product. The model has been developed mainly in the context of providing an account of written texts, so this development is hardly surprising.
 This confuses textual relations within a text (the concern of RST) with logogenesis: the unfolding of text at the instance pole of the cline of instantiation.
Moreover, if 'regressive dependency' were a problem for RST, it would also be a problem for Martin's 'logical' model, as the following quote (1992: 240) makes clear:
Conjunctive structures have been treated in other words as typically anaphoric, as reference chains. Conjunctions do sometimes point forward however.
See the (20/8/15) critique of this here.
 The suggestion is, more importantly, inaccurate, based, as it is, on the confusion of textual semantic relations with logogenesis. It would perhaps be unfair to suggest that the assessment is more hypocritical than critical.