this doesn’t solve the “codifying good games” problem but it’s a very small step forward for procedural generation of entire games: http://www.doc.ic.ac.uk/~sgc/papers/browne_cig11.pdf
it’s pretty dense and full of maths but basically the two authors created a system that acts as almost a markup language for defining game rules and then perform various operations on those game rules using selection criteria to filter out games that don’t work over tons of iterations.
the problem with generating games this way (and procedural generation in general, honestly) is those initial assumptions – in order to build a system that can even be iterated over and selected for, the authors of the paper put some pretty harsh constraints on what constitutes a “legal” game, and the games that are created aren’t interesting in any modern sense