For example, the alphametics exercise is listed as easy. This is not an easy exercise by any measure, as can be seen by the tiny number of submissions it has, and the fact that the person who ported the exercise to racket needed to come here to ask for help.
The sublist exercise is also pretty difficult, because the list lengths in the test cases prohibit some simpler, regex-based solutions and force performance optimization. I’d rank it at least medium.
On the other hand, the “Robot” puzzle is pretty easy.
It seems most of the exercises are at a rating level 1. While ratings are largely subjective and can get complicated fast, it seems at least some of these should be tweaked since it’s unlikely all those exercises would be the same difficulty. You could start a PR with some preliminary ratings and then folks can discuss further if so inclined.
If you go the bottom of the track’s build page, the Usage Statistics sub-section in the Create Practice Exercises section shows the completion percentage. This could be helpful.