How do you solve a problem like scholarly peer review?