Yeah, that's part of the reason I list the number of attempts as part of the stats for each model + respective prompt. It's a loose metric of how "steerable" a given model is, or put another way, how much I had to fight with it before we were able to get it to follow the prompt directives.