> end-to-end evals keep the harness honest sorry, but the README shouts: i am slop. no matter how good the code is. what is a honest harness? i would prefer at least a readme written by hand, im okay with ai generated code, but the readme is for humans not agents