Why do sloppy phrase queries return such different hit counts?

[mikemccand]

The two engines sometimes produce wildly different results -- how come?

E.g.:

"web url"~4   428 docs (Tantivy)   131,414 docs  (Lucene)

[PSeitz]

The phrase query handling with slop in tantivy 0.19 doesn't allow reordering of the terms, e.g. "A B"~2 matching "B A". This will be allowed with tantivy 0.20 and should match the behavior of lucene (quickwit-oss/tantivy#2020)

[mikemccand]

Aha, transpositions! Lucene's FuzzyQuery went through a similar evolution :)

Thanks @PSeitz!