It works. But the question is what you aim for:

• least lines of code
• most robust solution that you can easily reason about
• efficiency in terms of intervals
• efficiency in terms of the total length of intervals

I'm not saying one is definitely better than the other, especially since this is not a code for production, but for practice. For this reason it could also make sense to supply several implementations for each of the above goals.

just because the kata is too old to have proper performances tests.

But it works

but especially bad about performances...