Back in the stone age when I did mine, my approach was more local, but similar. I wrote all the text in LaTeX using AucTeX package, but had a makefile and a co

Back in the stone age when I did mine, my approach was more local, but similar. I wrote all the text in LaTeX using AucTeX package, but had a makefile and a commit hook (to CVS, later SVN, this was pre-git). All figures & tables etc. were generated from other make targets, which chained back to a mix of (usually) lisp code, matlab, or maple/mathematica. Oh, and figures generated by things like MetaPost. Structuring things at the chapter level .tex files made it all pretty manageable. I had figure dependencies on both input data and executable targets/scripts, so changing either would trigger a recompute, with the exception of some long running stuff - of course had global rebuild rules. I wrote many papers this way too, it was nice to know everything was synchronized properly. I did use org-mode to organize my work though. These days it would be on github and I probably wouldn't have lost it in a move.