Bloomfilters intentionally can have false positives but no false negatives. They're great for searching over very large datasets as a quick and cheap way to drop lots of data from more expensive search operations.

We use a deterministic checksum for this so you know what you're getting without ambiguity.