quantifying-consistency-and-information-loss Quantifying Consistency and Information Loss for Causal Abstraction Learning