That's really not obvious to human players even at a high level.Of course there's lots of scenarios where it's better to ignore that rule and that's what makes it really hard to make an AI, but I did find it interesting and I keep it in mind in arena now.
It would know things that humans can't know, like the win% increase that happens when you play X card at X turn.I remember someone tracked the win% of games in arena depending on what you used the coin for and it was significantly better to use it to coin a 4 drop.Finally, i really want to know why life couch spend some much time on every turn. In return for my gift of power, you must grant one of my wishes.Does he predict and stimulate opponent move and even think about the solutions to what may happen 3 turns later? If you enter this contract, you will live as a human, but also as one completely different. He takes his time to think through his plays for this and the next turns.So I rambled a bit much and probably repeated myself ones in a while so the TL; DR why a perfect HS bot wouldn't work: That been said, if a group of bright minds would come together and program the best bot they could, it would probably still beat 50% of the HS player base if it's a deck like face hunter, maybe Zoo and maybe druid.
That's mainly because the biggest part of the HS community is not that good at the game.
I see that Kripp also knows this as he will almost never ever coin a 2 even if he has 2-2-3-4 in hand.
This is the kind of information that a bot could gather for you, if nothing else. Last thing they want is a software that can beat every player (and as such be easily detectable). Very early bots have already beaten pro players on twitch.
1000 games in order to balance out the influence of randomness that is inherent to any TCG.
As soon as you know what cards are in your opponents deck the amount of calculation needed is fairly small and therefor doable.
The old shaman bot that used to run rampant on ladder was very good.