Comparing audio repair examples fairly

I would like to build a small workshop exercise around audio repair. The aim is to help participants describe what they hear before they decide which version they prefer, rather than treating every change as an improvement. I have a generated passage in mind as the source material. Keeping the excerpt constant would be the first rule. A page discussing ai music artifact removal gave me a starting point for the topic, but it does not settle how to structure the exercise. I would match playback levels before presenting the versions. Participants could write down observations before seeing labels. The original excerpt needs to remain available throughout. The worksheet could distinguish distracting texture from lost musical detail. We could repeat the excerpts in a different order and discuss disagreements, since a single preference vote would reveal little about the reasons behind it. What listening criteria would you add to this exercise?