well, its recursion size is O(log n). i don't think it's too many.

You are right.
Broken codes which raise exception didn't fail tests before.
This is due to a bug in exception handling.
This bug has been fixed now, which results in the Test cases to run as intented.
Thank you for pointing out!

+1 for downvote. I've seen a lot of inappropriate "best practices".
I think it's a really shame.

That doesn't cover the case. The case like testing ["arp", "arp"] with ["lively", "alive", "harp", "sharp", "armstrong"] is needed.
Or you could add an constraint of a1 in the description.

javascript, ruby, c#, clojure, java also?

kumited to coffeescript

It would be nice if I can get honors when I complete the same Kata using different languages.

It seems that broken codes which raise exception doesn't fail these tests.

This doesn't work for really large numbers (greater than 2,147,483,648 * 3).
This happens because bitwise operations convert it arguments to 32-bit integers.

+1 for 'New solutions should generally sort to the top'
Really good new solutions are at the bottom of page and nobody upvote them.
It would be no problem for old solutions because good solutions are upvoted and stay at the top.

i don't understand why this answer is voted as one of the best practices even if this doesn't have 'var' for i.

this doesn't work.
it returns true with [1].sameStructureAs([[2]]).