Most likely, the person asking this question does not know where to look for documentation. With 6 schools, 300 blog posts, 80 wiki pages, 500-page manual and 50-page tutorial, there is plenty of information for somebody willing to learn about Madagascar or to understand how it works.
Reproducible papers and books link scientific descriptions of different algorithms with examples of their usage and thus represent the best kind of documentation scientific software may have. On the last count, there were more than 500 computational recipes (SConstruct files) and more than 5,000 reproducible figures. Reproducibility means: we provide reproducible examples of using a particular program but we do not guarantee that the program will work properly with a different choice of parameters or with different input data. With time, the ecosystem evolves. Some programs are getting used more often and, as a result, are getting better debugged and more thoroughly documented.
Madagascar has a low barrier for authors to start contributing their work. When some of the newly contributed programs and examples are not sufficiently well documented, documentation becomes a process of communication between authors and users. Engaging users as active participants in this process will help us make it more efficient.