Using a 0-100 scale is a pretty bad idea in general. Neither humans not models know what to do with it. Split your metric into several concrete variables with discrete values well understood by the user (human or model). If you still need something more granular after that, calculate the aggregate score.