In the Crystal translation, all the sample tests use Array(String), while all the random tests use Array(Char)

While this doesn't matter for some algorithms, it will break others. With the invisibility of the random tests, this is very confusing.
I think the intent of the Kata is not to deal with different types in the language, therefore I feel this introduces unnecessary complexity unrelated to the intent of the kata.

Though I think normalizing it to all Array(Char) makes the most sense, the most important part is that the sample and random tests should use the same type.

(Though if it's important for the author that the algorithms should be able to handle both types, I think the sample tests should include both types as well)