Really underrated Kata, and was quite satisfying to solve, but there are a couple notes:

1. The range class provided is underbaked. Bare minimum, it should be default-constructible - meaning you need to add two braces to turn the declaration into a definition. Also, not exposing conveniences like range::from and range::to adds artificial complexity to the task.
2. The tests are in severe need of ExtraMessage in the assertions. Unlike most Kata, this one is not function-based, meaning the solver has no way to easily log things. It won't be easy, but you'll need to roll your own custom stringizer that converts types into string representations. I've done something similar in the preloaded section of this Kumite a while ago.
3. (Subjective) I personally am not a fan of how the reversed ranges are treated. Why [N + 1, M]? Usually, reverse indexing stands for [size - N, size - M] (at least in my experience). But this is subjective in the end, so ignore it if you wish

Thanks for the issue report, added random tests now.

Random tests are needed.