Challenging! Especially the possibility of a key's being int or string makes it even more compelling. Nice kata.

In the task n is said to be in range [3, 30], but random tests have examples with n==1.

In my opinion "a range of numbers" is an ambiguous description. Your examples show a range of consequitive numbers, and the Kata you are extending also uses the word "consequitive." It would be great if the description made that clear.

Ahh you're right :) my mistake

This is okay, because of this part of the kata description:

The terms of a subsequence need not be consecutive terms of the original sequence.

There's a small mistake in the description section of the python translation of this kata:

Example subsequence

Subsequences of "abc" = "a", "b", "c", "ab", "ac", "bc" and "abc".

"ac" is not a subsequence of "abc" :)

Lack of small preloaded tests, especially tests on arrays.
Debugging huge random tests is quite a pain.

see above (manual notification)

Hey :)
I've added random tests now - sorry about before I really misunderstood what was meant by random! Let me know if you think the test-cases help the strength of the kata - I'd love to get it published. I can keep investigating more efficient solutions too.

what about the possibility of having a list as first layer? (note: it would need a slight change in the description)

I just looked at the test suite and...

x/

those are not random tests, they are fixed tests. => look at some others katas you solved (recent one if possible) and study how the random tests are done.

Note: in your case, you'll have to compute the expected solution before you send the data to the user (or you'll need to make a deep copy of the structure)

{..., '2': []} should equal {..., 2: []}

This behavior isn't specified in the description (and is highly arguable): the original key was an int, but it hasn't been converted to a string in your solution. For now it's undescribed, but moreover, that sounds inconsistent to me (because if the first key is an int and you find another layer below, it will be cconverted to a string. So for homogeneity purpose, all keys in the output should be strings).

Side note: currently, the way the test suite is splitted (is it even splited..?) isn't correct: everything appears in a "random tests" block. => ?
Some readings you might find useful: https://github.com/Codewars/codewars.com/wiki/Codewars-Python-Test-Framework-V2