Thanks for sharing, John; this looks interesting! Would be nice to build a codec using that reference impl (hmm, except it is C++)

FST in this context stands for "Fast Succinct Trie" not "Finite State Transducer" and it seems to be a data structure similar to (but claimed better than) bloom filters, in that it can efficiently evaluate set membership with a tunable chance for being wrong (false positive), but it'd be nice to see some numbers on a real Lucene index / use case.