Opened an issue at the repository but I think the problem is somewhere in the HR-silverfish-interface. We will see.

I agree 100% that default AI is more powerfull than silverfish but silverfish is a lot faster.

Why is that? I'll try to explain. ;)

At the very start there was DragonCC using an editable rule set for card plays (eg. use card Fireball if - and only if - minions health <= 6 and attack >= 4) which was very situational and hard to handle.

Then Silverfish invented a value based system using an implicit and hard-coded rule set for evaluation.
This has been a great improvement and doesn't require a rule for every situation anymore and rules are only checked if they are required.
The downside is a change in behavior requires a rewrite of the evalutation function.

Right now we have the HR default AI, which combines both advantages of the value system and the editable "do and don't" rules. Thus, it is more powerfull than both.
The customization on the other hand makes it a lot slower than it could be because it has to check every single rule and can't combine rules or ignore them if they are false anyway.
This is because of the OR conjunction of all lines. Even if one more line doesn't seem a big deal at the first sight you have to keep the huge ammounts of boards calculated in mind.
IMHO the key to a fast default AI is to add as few lines as possible and having a separate ruleset for every class/deck.

The next generation of AI is either learning (writing its own rule set over time) or cooperative (what is current meta at rank X, what cards has player Y in his deck, what's the playstyle,...).
Perfect would be a combination of both. ;)

Hardware improves and the number of boards calculated can be reduced, so unhandable is a big word and I think you can still throw in a whole bunch of new rules. ;)

Perhaps JoyAdmin can give us better insight in default AI implementation and its limits.

I guess the CAST IF part should look something like:

if(expr){ "do something"; }
with:
expr = row1 || row2 || ... || rowX;
row = column1 && column2 && ... && columnY;

This means expr is true if any row is true and a row is true if all columns are true.


The GAIN VALUE FOR part is far more complex and could look something like:

for(any row in rules){ if(row){value += "gain for row"; }}
with:
row = column1 && column2 && ... && columnY;