Ciphers don't have to be consistent.

That's a challenge. In a one way hashing, say for passwords, the hashing is going to be consistent as you will want to compare a possible answer for a password, after a hashing, to the hashed value for a password at a database. If a different data point is merely encrypted, say a social security number, well that is a different thing. 457-55-5462 could look like two different things when encrypted as the cipher to dencrypt can handle as much, but this shape of things, a shape never intended to do the hash comparisonesque comparisons comes with some challenges. What if you want to grab a record by a SSN? You will have to query all the records, perhaps dumbing the records down to just id and SSNs alone, and then loop through the records doing an unscrambling before finally fishing for what you need in the in-memory table.