Maybe solving this would be determining a set of likely encrypted messages.

They could be scored by decrypted content matching structure of priors and context. Along with char difference from the encrypted message on record.