RMarkdown is good stuff. Hopefully, you've already heard of RMarkdown. If not, you'll understand it pretty quickly by looking at this simple example:

http://rpubs.com/medined/replacing_part_of_time_series_using_time_based_selection

Essentially, you mix R code with Markdown markup to create a 'living' document. The R code is executed when the page is displayed. The full power of R (and all of its extensions) can be used. There are many examples of this online.

RMarkdown pages can be computer-generated. Imagine if any given analytic documented the intermediate steps from Data Load to Final Visualization using RMarkdown? I bet user confidence in the final product would increase. It would also be trivial for the document to be duplicated and tweaked (draft mode) before being republished. Since RMarkdown is text-based, you could provide human-readable diff reports between analyses. Another advantage of this text-based system would be full-text search across all analyses.

I could also point out the value in being able to produce an analytic report without needing to know Java, Python, or another programming language. Just knowing the math is complex enough.

Update: http://studio.sketchpad.cc/sp/pad/view/ro.9QNw0rsxwki4J/rev.480 - The archive timeline widget allows visitors to view all versions of the source document.

Update: You can do this same kind of thing with Python code. Check out http://ipython.org/notebook.html.

I'm not saying R is the answer to all problems. But this idea of archivable, diffable analytic solutions was interesting.