A more realistic test would require, e.g. encrypting a large number of potentially somewhat structured inputs... and then decoding some of them. Weaknesses lik

A more realistic test would require, e.g. encrypting a large number of potentially somewhat structured inputs... and then decoding some of them. Weaknesses like that can be translated to real exploits in usage. Unfortunately, doing that would leak information the structure of the backdoor based on which texts could be decoded ... so you'd want to isolate that knowledge.