calling it an AI failure when the human maintainer made the same mistake says more about the difficulty of the problem than the model

Kind of insane that anyone would think AI can only patch things correctly 26% of the time