[derek73]

Closes #1080

What moved

Before a family comma, a one-letter roman numeral read as the generation from 2.7.0 on wherever a particle stood in the part, the particle counted as a name word to spare:

Input	1.4.0 – 2.6.0	2.7.0 – 2.8.2	this PR
SMITH VD V, JOHN	family SMITH VD V¹	family SMITH VD, suffix V	family SMITH VD V
Smith van der V, John	family Smith van der V	family Smith van der, suffix V	family Smith van der V
van der V, John	family van der V	family van der, suffix V	family van der V
de Mesnil V, Jean	family de Mesnil V	family de Mesnil, suffix V	family de Mesnil V
John van der V, M.J.	family John van der V	family John van der, suffix V	family John van der V

¹ 2.2.0 and 2.3.0 read family SMITH V, suffix VD (#573).

Why

rules/C1's numeral site in the family part (_assign._family_part_numeral, added by #886) asks rules/S10's (the trailing roman numeral) words-to-spare count: a one-letter numeral needs two name words in front of it. It counted pieces, and a particle no chain has joined to a name word is a piece of its own there. _names_before now counts only the pieces that are not wholly particles: the comma has named the part a surname, where a particle heads the name word behind it, so it reads as part of that word, as rules/C3 (counting the words before the comma) counts a particle run.

Rules amended

• S10 (the trailing roman numeral), family-comma Precedence bullet: states the count, with Smith van der V, de Mesnil V and the Smith van der Berg V boundary as examples. The clause reads the signal S10 already reads there (its words-to-spare count), so it is not a new rule.
• C1 (the suffix comma decision): its sentence that a numeral ending the family part "is read as the comma-less name reads it" gains the exception, with John van der V, M.J.. The comma-less name has a given name a particle may follow; a family part has none.
• P9 (a particle that is also a suffix word, beside a family comma): the deviates: #1080 markers are gone, the clause is scoped to a letter ending the part, and its mixed-case example (Smith vd v, John) is now fenced.
• Rule sizes re-recorded (rules_budget.json): C1 454 → 476 words, P9 405 → 420, S10 665 → 721, each for the clause above.

Boundaries (case rows)

• Smith van der Berg V, John keeps suffix V: a joined chain is one name word, so two stand in front, as in Smith Jones V, John.
• Smith Jones van V, John and de Mesnil Jones V, Jean keep suffix V.
• Numerals of two or more letters, the comma-less John van V, and the part before a suffix comma (Smith van der V, Jr.) do not move.
• Not decided: with a suffix word behind the letter, P9's first clause reads it as a name word (Mary Smith vd V Jr., John gives family Mary Smith vd V), where the part without the particle reads it as the numeral. It reads the same on master and is recorded in decisions/P9.

Measured

• Grid of 34,992 parses (recipe in decisions/P9): 4,398 move, 1,466 per order. Every one is a one-letter numeral moving from the suffix to the family with one fewer suffix-or-name.
• Committed corpora: only SMITH VD V, JOHN and smith vd v, john move, plus five rule examples new to the rules corpus.
• Local corpora against a 401bf62 checkout: ctgov-officials and tse-donors do not move. fec-individuals moves two: DU HART V, JOHN (suffix V → family DU HART V) and VAN METER V, ISAAC C IV (generation V, IV → family VAN METER V, generation IV). In FEC's FAMILY SUFFIX, GIVEN writing both Vs are most likely generations. It is the cost S10's count already takes for a one-word surname (SMITH V, JOHN reads family SMITH V on master), and it is 1.4.0's through 2.6.0's reading. Decided: accepted (decisions/P9's Why does "SMITH VD V, JOHN" lose the V from the family name when "SMITH V, JOHN" keeps it? #1080 entry). The first draft, which left out only particles right in front of the numeral, moved neither.

Checks

uv run pytest, uv run pytest -n0 -m serial, uv run mypy and uv run ruff check pass. Reviewed by the design-docs, code, test and comment reviewers; the second commit addresses their findings. A second round (code and design docs) reviewed the fix commits: it found no code defect, and the last commit files the count under decisions/S10, adds head notes to the two entries that restated the old count (decisions/C1 #774, vocabulary-model 2f), and corrects three release ranges.

Not done here: the connective twin Smith y V, John (suffix V, family Smith y V in 1.4.0 and 2.6.0), which the test review found and which is a separate rule, now #1093.

🤖 Generated with Claude Code