Read a lone V, I or X after a first name as the family name (#718)

[derek73]

Closes #718.

What changes

A single capital I, V or X ending a name written without a comma now needs a word to spare, counted exactly as rules/S4 (the ambiguous acronym needs words to spare) counts Ma: in name units, a particle chain counting as one word and a title as none.

input	2.4.2	now	1.4.0
John V	given John, suffix V, no family	given John, family V	last V
Malcolm X	suffix X, no family	family X	last X
Dr Smith V	family Smith, suffix V	given Smith, family V	first Smith, last V
John Smith V	suffix V	suffix V (unchanged)	suffix V
John W. Ingram V	suffix V	suffix V (unchanged)	—
John Smith vi	suffix vi, reports	suffix vi, reports nothing	suffix vi

• Reports. Either reading reports suffix-or-name, both ways as Jack Ma does (Derek's call), from one site: the declined numeral stays one of the trailing read's picks. The one exception is a letter that classify already reported as an initial (HENRY I, one-case i), which reports nothing new.
• Lower case. A lowercase numeral in a mixed-case name is settled by its writing: it reads as the generation and reports nothing.
• Comma paths. The given part after a family comma is not counted, as S4 already says of the acronym.

Docs

• New rules/S10 (the trailing roman numeral) and decisions/S10 record the decisions, what moved, what was declined and why, and the recompute recipes.
• Amended rules:
  • P2 (the chain's stop, and its boundary example Freiherr von Richthofen V)
  • P3 ("the generation it also spells" now defers to S10)
  • P5 (the words-to-spare list)
  • M2 (examples)
  • O5 (its worked example is now John III, Dr.)
  • S2 and S4 (implemented:/interacts:)
• Elsewhere: a release-log bullet, and two AGENTS.md amendments: the spelling sweep, and the group-side-predictor gotcha.

What moved (snapshot diff)

Four pre-existing corpus names:

• John V → family V
• Mohamad X → family X
• John née Jones Smith V → maiden 'Jones Smith V', since the clause-free name John V no longer reads the V as a suffix
• Freiherr von Richthofen V → family 'von Richthofen V': behind a title the chain is the one name word, the same reading Freiherr von Richthofen Ma already gets

The snapshot also gains S10's examples as new names. Re-pinned case rows are listed in decisions/S10; the two whose purpose needed a suffix kept their purpose on III spellings, and the old inputs became new rows.

Review

• Design-docs review (eleven axes) found 8 issues, all fixed in 98f75f5. The main ones: S10's Accepted clause over-scoped to i; P3 and O5 were stale; the release bullet missed Dr. Smith V.
• Code review found that the count leaked into a family comma's given part through the chain's bound: Smith, Freiherr von Berg V read given 'von Berg V', unreported, across 2,880 grid parses. Fixed in ede4d25. A second pass re-ran the same 558,423-parse grid over the fix commits: 0 unreported numeral moves remain (2,880 before), every reading ede4d25 changed equals master, and the flag introduced no new divergence.
• PR review (comment accuracy and test coverage, which removed each emitter in a copy of the tree to see what failed). The fixes are in the latest commit:
  • Fewer report sites. The particle chain's numeral report only chose a wording, since assign reports the same kind from the final role, so it was deleted. The maiden clause's stand-down could never fire and was deleted too.
  • The clause now checks case. A mixed-case lowercase numeral given up by a maiden clause reported where every other path was silent (Jane Doe nee Smith vi). It now asks the case, as assign does.
  • Labels. Six classification labels were corrected against the 1.4.0 wheel, and five comments made stale by the change were fixed.
  • New rows pin each emitter, stand-down and gate. Each mutation now fails at least one test.
• /simplify (reuse, simplification, efficiency, altitude): behavior-neutral cleanups in 4ac5a06. The lowercase case decision is now made once, in the walk; the one-letter test reuses is_single_letter_numeral; and reported_as_initial moved into _assign. Efficiency was measured: at most +1 frame, only on names ending in a numeral.
• Follow-ups filed: Should Elizabeth II, Queen read Elizabeth as the given name, the way Queen Elizabeth II does? #720 (H1 behind a comma) and Should john v read V as the family name, as JOHN V and john x do? #721 (john v).

For Derek

1. Freiherr von Richthofen V now reads family 'von Richthofen V'. That's consistent with S4 (... Ma), but it is a corpus name.
2. john v (one-case, lowercase) still loses its family, while john x and JOHN V keep theirs. A lowercase v is a listed suffix word and never reaches the fork. This is recorded as Accepted in S10; closing it would mean touching the suffix-vocabulary path.
3. H1 behind a comma is now unpinned. John V, Sir was the only spelling where a given-name title behind a comma kept the lone word as the given name (H1). John III, Sir, John Jr., Sir and John PhD, Sir read family on master already. The row is renamed for what it pins. A follow-up issue is worth drafting.
4. Garbage-input observation: Berg nee MA V used to report the clause's kept MA; now the clause keeps MA V and reports only the V.

Checks

• Tests: the full suite (21,126 passed), mypy and ruff are clean.
• Frame budget: test_benchmark passes.
• Snapshot: re-recorded.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.03%. Comparing base (958d252) to head (4ac5a06).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #719   +/-   ##
=======================================
  Coverage   99.03%   99.03%           
=======================================
  Files          46       46           
  Lines        4557     4575   +18     
=======================================
+ Hits         4513     4531   +18     
  Misses         44       44           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.