Reproducible documents are a reader interface for research and their implementation does not require a specific choice of computational tools and conventions. A simple black-box description of input and output files is all that is needed: Any result is usually represented by a single output file (usually a data file). The result file might have various display versions such as a postscript file or Gif file that are all trivially derived from the output file. A command (which could be a composite of several program invocations) computes the output file from a set of input files. The input files are usually data files and programs. I refer to the set of programs that compute the result files from those source files as the application. Note that the application may very well include the compilation of programs.
In the past we considered the data result file an intermediate file. Today, I feel that this data result file is the actual result and that the various graphics representation are trivially derived from it. By making the data result file central, the reproducibility becomes far more flexible with regard to various output formats such as postscript, dvi, or HTML documents. The shift in philosophy is important and has various practical consequences.