"Is built fairly" is easy to define, and means: doesn't collude with other players. I mean, that it must not say: "if the higher-ranked player (or the most handsome, or the most Dutch, whatever) is in the lead, buy Province". It's obvious but worth mentioning.
As long as it's as likely to screw you as it is to screw me, then I'm fine with it.
If I'm playing three player, I must be ready to get screwed - by bot or by human.
Of course you'll want a better bot than our Rattington in that position. But you'll want a better bot than Rattington in any
I feel like assigning leaderboard results to this kind of thing requires that you have a bot that isn't going to do stuff like this, ever.
But humans do stuff like this all the time. I play a good deal of 3-4p IRL, and I will take the last Province to ensure a second position, or push the piles to dangerous levels to get something I really need to make an unlikely comeback. In short, I will interfere even if I'm not in close contention.
So why set an extra-high bar for bots?
I mean, I can see your argument working against 3p dominion in general, but I don't see how it works against bots in particular.
As a rescue mechanism, I believe that replacement by bot wouldn't be problematic in the vast majority of games - even using Rattington would maybe be ok in the meantime, but I'm less sure - I haven't played Big R in a long time.
It's just a different thing in my mind to have a human player do this and to have a bot do this -- you can blacklist that human, or you can expect to climb high enough above them on the leaderboard to not have to play them anymore, but the bot will always be there.
Well, the bot can always get better. We can surely agree that your example is very extreme. An OK bot wouldn't do that, and an OK human wouldn't do much better than the OK bot in terms of non-interference.
It may be a slippery slope, but then you're already sliding when you start a 3p game.