I mean baselines to match how the LLMs are being tested in whatever paper (as best as possible). In this case, average scores on the unaltered benchmark and the

I mean baselines to match how the LLMs are being tested in whatever paper (as best as possible). In this case, average scores on the unaltered benchmark and then average scores on each modified benchmark to indicate how much on average human performance drops introducing these details. A "hey, you got that wrong. check again" is fine if the LLMs in the paper are also being prompted that way.