A leading suffix displaces a particle out of leading position exactly as a title used to before #367, so the chain fires and the name ends with no surname:
Ph. D. Van Johnson given='Van Johnson' family='' suffix='Ph. D.'
That is the same defect #367 was filed for (Sir Van Johnson → given Van Johnson, family ''), reached through a suffix-shaped leading piece rather than a title-shaped one.
Why #367 fixed only the title half
The transparency scan steps over a piece that can only be a title. Suffix pieces are deliberately not skipped, and the reason is real but was recorded backwards in that PR (corrected in #370): skipping them moves Van out of the family and into the middle for the shapes whose leading piece assign puts in given:
shipped suffix-skipping variant
Ph.D. Van Johnson given='Ph.D.' family='Van Johnson' given='Ph.D.' middle='Van' family='Johnson'
II van Gogh given='II' family='van Gogh' given='II' middle='van' family='Gogh'
Esq. van Gogh given='Esq.' family='van Gogh' given='van' family='Gogh'
So a blanket "skip suffixes too" trades one defect for another. Ph. D. Van Johnson gains a family name; III van Gogh and Esq. van Gogh get worse.
The distinction that matters is what assign does with the leading piece: Jr. Van Johnson is already correct because jr is in TITLES and the piece is a title. The broken shapes are the ones where the leading piece is suffix-shaped but lands in given.
Measured scope
The suffix-skipping variant is a no-op over all 751 differential corpus names and passes the full suite, so there is no baseline evidence either way — the decision rests entirely on which of the shapes above matters more.
Suggested framing
Rather than skipping suffix pieces wholesale, the question is whether a leading piece that assign will not treat as a name should participate in the leading-particle exception at all. That is a narrower rule than either alternative and would need its own measurement.
Pinned by tests/v2/pipeline/test_group.py::test_a_suffix_shaped_leading_piece_is_not_stepped_over as of #370, so a change here announces itself rather than passing silently — before that test, the mutation flipped this decision with a green suite.
Split out of #367 so that PR stays one mechanism.
A leading suffix displaces a particle out of leading position exactly as a title used to before #367, so the chain fires and the name ends with no surname:
That is the same defect #367 was filed for (
Sir Van Johnson→ givenVan Johnson, family''), reached through a suffix-shaped leading piece rather than a title-shaped one.Why #367 fixed only the title half
The transparency scan steps over a piece that can only be a title. Suffix pieces are deliberately not skipped, and the reason is real but was recorded backwards in that PR (corrected in #370): skipping them moves
Vanout of the family and into the middle for the shapes whose leading pieceassignputs ingiven:So a blanket "skip suffixes too" trades one defect for another.
Ph. D. Van Johnsongains a family name;III van GoghandEsq. van Goghget worse.The distinction that matters is what
assigndoes with the leading piece:Jr. Van Johnsonis already correct becausejris inTITLESand the piece is a title. The broken shapes are the ones where the leading piece is suffix-shaped but lands ingiven.Measured scope
The suffix-skipping variant is a no-op over all 751 differential corpus names and passes the full suite, so there is no baseline evidence either way — the decision rests entirely on which of the shapes above matters more.
Suggested framing
Rather than skipping suffix pieces wholesale, the question is whether a leading piece that
assignwill not treat as a name should participate in the leading-particle exception at all. That is a narrower rule than either alternative and would need its own measurement.Pinned by
tests/v2/pipeline/test_group.py::test_a_suffix_shaped_leading_piece_is_not_stepped_overas of #370, so a change here announces itself rather than passing silently — before that test, the mutation flipped this decision with a green suite.Split out of #367 so that PR stays one mechanism.