Nonsense, the kata is fun to the point when you start encountering weird ideas that author came up with to "spice" things up. Most ridiculous of which is something called "rollback", which I have no idea how it is supposed to work and description certainly does not clarify this. It has nothing to do with the game of Go and is total nonsense. Basically I have written the whole solution but cannot get it through random tests because idk what author had in mind.

In theory I agree, but as a user it's frustrating to spend half an hour to an hour figuring out the solution to than have to give up because you cannot find some specific optimization. For me and probably most of this site users it's unclear if the Kata will be 90% problem and 10% speeding it up or the other way around. Also I don't think the said approach with 'counter' works with all implementations of tests although I might be wrong. From the user experiance perspective the lack of corralation between idea quality and execution quality of Katas is the most conspicuous thing about this site. Anyways thanks for the reply.

I would very much like if something was done with performance Katas. Right now the situation is that you can find interesting Kata, solve it and not be able to submit it due to not good enough speed. As far as i know there exists no way of actually telling how far your solution is from being good enough which makes improving your solution annoying. Also provided test often do not include performance versions of the problem. This means you really just cannot test your code on a problem. To add to this some languages are faster than others and I found examples where you can brutforce solution with C but not Python. I get it, it's hard to manage all this in a right way.
I don't think however this should be the reason for sloppiness.

My sugestion would be to make performance and normal version of all those performance Katas or at least flag them. Secondly it should be clear what kind of performance you are getting and what is expected. Otherwise all those problems are basically training in Googlefooing the most efficient algo.