We address the problem of improving, automatically, the usability of a large online document. We propose an adaptive hypertext approach, based on splitting the document into components smaller than the page or screen, called noogramicles, and creating each page as a new assemblage of noogramicles each time it is accessed. The adaptation comes from learning the navigation patterns of the usors (authors and readers), and is manifested in the assemblage of pages. We test this model across a number of configurations, including chance and non-adaptive systems. We evaluate our model through simulation. We have designed a simulator based on established findings about the behaviour of hypertext users. We have realised a quantitative evaluation based on hypertext usability measures adapted to the problem: session size, session cost.