Like I said, I don't believe Goko is using the set generator anymore, but it would give me peace of mind (or something) to know for sure.
It seems easy to look at the code; maybe it's not some secret server-side thing. Wei-Hwa's thing is its own file (checks), dominionSetGenerator.js, and anything using it will have to say (checks) DominionSetGenerator. So search for that string in other files. Probably you will find it commented out somewhere.
Casual and Unrated games use the set generator, pro games do not. (At least, the last time I checked the code a couple months ago, but I doubt it's changed since then)
Interesting... that would explain why I never see certain cards when playing bots (like Feast) but constantly see some (like Philosopher's Stone). It's subtle, but there are definitely some cards getting precedence in casual.
I did some unscientific sampling, and I think this is just confirmation bias. I took 10,000 random pro games and 10,000 random casual games. Then I removed all the promos and base cards (since I figure these will show up with frequencies that relate more to sets owned than anything else). For each card, i counted what % of pro games it showed up in, and what % of casual games. Then, I divided these two percentages by each other, and sorted them, to see which cards are most heavily weighted to pro and casual.
Every card is within 30% of "normal" - that is, showing up exactly as often in pro as in casual. A couple of the most "fun" cards (KC, GM) show up only 72% as often in pro games: one reasonable interpretation of this is that when someone pre-sets a casual game, they include these most often. Likewise, a few of the least fun cards (Cultist, Pillage, Mandarin(?)) show up 25-30% more often in pro games; probably, pre-set games don't include these cards because nobody loves seeing them on the board.
I'm not sure what the "expected" deviation would be for a perfectly random card selection, but 30% seems pretty plausible - I think an automated "introduce the fun cards more often" process would be more biased than this. But, if anyone cares to run more statistics on this data, it's available at
https://gist.github.com/amalloy/faebd4509ad922a982d7 (where the numbers are obtained via the method described above).
As for jsh's observation about Philosopher's Stone: it appears 20% more often in pro games than in casual games. I wouldn't worry that Goko is trying to sneak more of them into your bot games!