I'm not sure how true that is; procedural games I think tend to come up with a core formula once, and stick to it pretty hard. The rest is just playing with numbers, and presets. Like for minecraft, there's a bit of complexity involved to it's procedural generation, but in general I imagine it hasn't changed much since its inception, with possibly two, maybe three, rewrites to its map generation logic.

Most of the "real" updates to minecraft are in regards to enemies, item recipes, "biomes" (which presumably don't change the generation logic much, just the spawn rates), etc.

Roguelikes I imagine are also in the same vein; it's all just fiddling with the numbers, for balance and "fun"

Simulation focused games on the hand are probably a better candidate for "purely programming" programs, but even then, the complex stuff is probably in the vein of emergent behavior; something you don't at all strive for in normal programming. And they still presumably make a large chunk of their decisions based on "fun" instead of reality. The harder you strive for reality, the less a game it becomes... Maybe Grand Strategies then; they pretend to simulate a pre-generated complex environment, but avoid emergent behavior for the most part.

Really the only thing I could safely call a "pure programming" problem is a tech demo that happens to be a video game, ie Carmack's stuff. But even then...

I think it might just boil down to that very few video games are both good, and disinterested in player enjoyment. The key distinguishing feature of video games as a media is human interaction; it's difficult to imagine a good game where this is not the predominant motivator. Even Dwarf Fortress keeps it in mind pretty strongly.