Hearthranger's one click and go arena with zero features and sketchy play has to be the weakest part of the whole bot. Makes tons of mistakes, is buggy, and has no customization or options of any kind. Arena is easily the best way to get both gold and cards in the game of Hearthstone, and it would seem the A.I. regarding its function needs a serious rework. So far I've seen little mistakes like not playing Knife Juggler first when bringing out more than one minion or not placing a buff card like Defender of Argus and Dire Wolf Alpha in between 2 creatures already on the board, or being too heavy-handed with buff minions such as playing 2 Abusive Sergeants when one is sufficient. I've seen medium mistakes like Mage pinging the face before removing a more powerful enemy creature with one life so that creature gets another turn to go face. Heavy, game losing mistakes like giving all of the tempo built up over to the enemy by choosing not to act when it clearly should have, 2 TURNS IN A ROW.

I know I should be grateful, and I am. The bot grinds like a beast and for the most part makes fair, if not efficient moves, and being able to walk away from the game and do things irl without babysitting it is pretty priceless. I just wish I had more say, short of grabbing the controls myself, in controlling how it plays my cards. Perhaps a feature that allows deckrules to be implemented like in the game settings with both ranked and unranked play. Then again Hearthranger is still in beta so I guess well just have to see.

This has been the case and unfortunately I don't think it'll ever be solved.

Making a good AI is 90% of the work in Hearthstone botting. It will probably take at least 100 hours to make a good Hearthstone bot that doesn't make any obvious mistakes. And let's say you make $1000 worth of hearthstone cards from it. That comes out to a measly $10/hour. Anyone who can do a decent job won't be working for minimum wage.

I've looked at the silverfish code, and no offense to the massive amount of work that botmaker has done, but I can tell he doesn't program for a living. I found a lot of inefficiencies and logical errors. And the hearthranger AI seems to be farther behind because joy needs to work on the hooking code.

No one is going to rewrite the simulations for 500 different cards from the ground up. And they need to be rewritten because right now it only returns 1 possibility per card which means you cannot accurately simulate cards like knife juggler or druid choice cards.