Hacker News Reader

My impression was there was a lot of research and competent results in StarCraft AI.

And that was 10 or so years ago.

Which is more constrained than your average 4X, but not fundamentally different (resource + research + movement + combat).

Not really. The AI was essentially "cheating" to play inhuman micro, its macro was apparently just as bad as typically expected.

I remember (from HN) some genetic algorithm explorations of optimal research/build orders that discovered novel and better combinations then-unknown to the playerbase.

That might well be the case, but it didn't translate to realtime play AFAIK.

It would be a major breakthrough if the AI could plausible adapt to the human players.

Every one gets a challenge (or not if they chose so). We are probably not quite there yet.

Doesn't Galactic Civilizations 2 qualifies ? (Maybe for not particularly high values of "experienced".)