Thanks for your patience. I asked Jason for more insight on Flukebook’s match scores and this is what he shared:
All three algorithms have relatively inscrutable scores. They’re not probabilities and not especially meaningful but are rather distance metrics of one form or another and will vary from trained instance of the algorithm to another trained instance.
PIE v2 can sometimes show a significant score gap between a strong match and a weak match. But the scores themselves are not meaningful.
We recommend that users focus on the ranking.