Is cryptography engineering or science?

Proving an implementation reliable, IMO, is impossible. We can only check them against known attacks, and surprising new attacks pop up every now and then. In the extreme, we cannot know about new physics before it is discovered (a dark matter reciever might be able to peek at encryption keys from miles away. If it did, our idea about how the universe works would change enormously, but it is not something we can rule out)

So, I think one can only prove the mathematics safe and there, AFAIK, we are far from any solid result. This article mentions "AES is known to be immune to differential cryptanalysis", but again, that is a proof for a known attack. In a related field, we don't even know whether one-way functions exist. In the end, there might be a result that shows that cryptanalysis can't be exponentially harder than encryption (unlikely, but AFAIK, not disproved)