Make sure when doing your decode that your key is using the "decoded" text and not the encoded text, meaning you can't pre-compute the key on decode.

From the description:
With the basic Vigenère Cipher, we assume the key is repeated for the length of the text, character by character. In this kata, the key is only used once, and then replaced by the decoded text

That test case is right. It's actually there to make it easier to conceptualize how the polyalphabetic shifts work in this particular Vigenere variant.

AAAAAAAAPASSWORDAAAAAAAA with the key "PASSWORD" and a standard uppercase latin alphabet should indeed result in an encoded string of PASSWORDPASSWORDPASSWORD, so it seems like you're misunderstanding the definition. Also, the test cases don't refer to any variable called key, so I'm guessing that's on your end.