Read a one-case one-letter numeral as its capital does (#721)

[derek73]

Closes #721.

What changes

A name written entirely in lower case now reads exactly as the same name in capitals does.

input	master	now	capitals twin
john v	suffix v, no family	family v, reports suffix-or-name	JOHN V → family V
john smith v	suffix v, silent	suffix v, reports	JOHN SMITH V reports
john j v	family j, suffix v	middle j, family v	JOHN J V → middle J, family V
john i	conjunction-or-initial	suffix-or-name	JOHN I (same)
doe, jane v phd do	suffix v phd do	middle v, suffix phd do	DOE, JANE V PHD DO
john i phd / John Smith V PhD	silent or conjunction-or-initial	same fields, reports suffix-or-name	JOHN I, PHD reports

• The tag. In a one-case name, classify tags a one-letter roman numeral as an initial, the same way its capital is tagged. Every reader that already treats V as an initial now treats v the same way. The numeral fork's test of the word in front counts a bare letter as an initial in a one-case name, as its capital would be.
• Your issue comment, and the PR discussion. A connective letter with no name word after it in its part connects nothing, so it no longer reports conjunction-or-initial (rules/P3 now states the exception). "No name word" means nothing at all, or only credentials and titles: john i phd, john i dr., jose e jr. Where the letter is a roman numeral ending the name, it reports suffix-or-name. That holds before a credential too, with the fields unchanged: John Smith V PhD keeps family V and now reports. A marked letter that isn't a numeral is silent (john e, and jones, john e, a corpus name).
• Should "John V" and "Malcolm X" read V and X as the family name rather than a generational suffix? #718's two-word count stands. Two stricter rules you weighed are declined in decisions/S10, with evidence. V and DVM #36's Hardie Leannon V and Problem parsing name with , V #136's John W. Ingram V both write the generation without a comma. The sheriff's own spellings, John W Ingram V and John Ingram V, are now case rows.
• One withdrawal, over final roles. Assemble removes classify's connective report from a marked letter when no given, middle or family word follows it in its segment. Assemble already withdrew this report kind where the letter ends up a suffix or title. It covers the no-comma name, a particle chain (john van e phd), the given part after a family comma (smith, john e phd) and an empty head. The numeral's report is emitted first, so one token never carries both, and Should "John V" and "Malcolm X" read V and X as the family name rather than a generational suffix? #718's stand-down is gone. The maiden clause reports a numeral it keeps before a credential (Jane Doe nee Smith V PhD), as the main name does.

Invariant

test_a_one_case_name_reads_as_its_capitals in tests/v2/test_properties.py runs a 1,872-pair grid. It covers every comma shape, all three orders, a maiden clause, a particle chain, a bare ASCII or non-ASCII letter in front, a trailing title, a doubled numeral, a trailing period and a shape-only three-letter numeral. It checks that each lowercase name matches its capitals twin in fields and report kinds.

• Master: 315 pairs disagree.
• This branch: none.
• Controls: one recorded negative control per guard: 225 pairs disagree with the tag off, 117 with the letter-in-front change off.
• Blind spot: a lower-vs-upper grid can't see a change that moves both spellings alike. The non-ASCII letter in front is therefore pinned by case rows, and the mutation that broke it fails them.

What moved (snapshot)

Six pre-existing corpus names:

• john v and doe, jane v phd do change fields.
• john smith v and larry james edward johnson v gain a report.
• john i changes its report kind.
• jones, john e loses its report.

One former rules example moves too: john nee smith i now reads maiden smith i. Docs updated:

• S10 (the trailing roman numeral): the lowercase-v Accepted clause is gone; the clause on a letter in front is extended; the stand-down's precedence over P3 is stated.
• P3 (connectives): a letter ending its part is not reported.
• S7: its example input is kept, with the new reading.
• decisions/S10 and decisions/P3 record all of this.
• AGENTS.md, plus the release-log bullet extended with :issue:721.

For Derek

1. The Catalan off switch (removing i from the conjunctions) no longer restores 2.3.0's three one-case readings (rovira, i and two others). With i no longer a conjunction, it's a one-letter numeral and reads as I does. The switch still unjoins Josep Carod i Rovira.
2. Two comma rows go silent rather than gaining suffix-or-name: rovira, i and maier, amy i, jr.. The comma path has never reported the numeral (decisions/C1).

Follow-up

• Should j nee smith read like J NEE SMITH? A lowercase single letter isn't an initial in a one-case name #723: in a one-case name, a lowercase single letter other than v, i and x still isn't an initial, so j nee smith reads family j while J NEE SMITH reads given J. This predates this PR and is recorded in decisions/S10.

Review

• Design-docs and code reviews ran on 5a6b7bb and agreed on two real defects:
  • The stand-down I had removed as dead wasn't: the trailing-title shape still reaches it.
  • The fork's test of the word in front was case-dependent (john j v).
• Fixed in ba7b25e: both of those defects, five record corrections, and two cost cleanups (a lazily built end-of-part set, and a C-level _ROMAN test), so ordinary names pay no extra frames. The reference row reads 366 frames on both master and the branch.
• PR review (a fix-commit re-review, comment accuracy, and test coverage by mutation). Fixed in 1f37d0a and 3d3b447:
  • The letter-in-front test overreached on non-ASCII and NFD letters. It now asks whether the letter's capital is initial-shaped.
  • Ten comment and docstring inaccuracies.
  • Rows for every unpinned decision the mutation run found.
  • The facade off-switch test is renamed for what it now pins.
  • Each new guard was then mutated on the final tree, and every mutant fails at least one test.
• Last review (code and design-docs, on 1f37d0a and 3d3b447). It found P3's broadened exception implemented only for a one-token piece on the no-comma path, with the stand-down still reachable. Fixed in 6e84cb1, together with the record corrections:
  • option B moves 19, not 18, with the recipe stated;
  • V and DVM #36 and Problem parsing name with , V #136 aren't shown to be real people; the sheriff is;
  • 1.4.0 parity holds for the capital spellings only.
• Rationale updated: decisions/S10 now says Should "John V" and "Malcolm X" read V and X as the family name rather than a generational suffix? #718's lone-numeral reading rests on the count, not on the generation being rare. Your searches showed generational markers are common, written without a comma unless a credential follows.
• Frames: every frame-check name matches master, including one-case names with a suffix.

Checks

• /simplify (reuse, simplification, efficiency, altitude), in 053f0e7. The "connects nothing" check had been answered in two places: classify's end-of-part set and an end pass in assign. It's now one check in assemble, with the same readings everywhere except one: jose e nee and smith, e nee smith report again, as master did, because the marker there is read as a name word. Removing the check fails 20 tests, and frames match master.
• Tests: the full suite (21,312 passed), mypy, ruff and the Sphinx doctests are clean.
• Snapshot: matches.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.04%. Comparing base (2bfa953) to head (f199dd3).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #722   +/-   ##
=======================================
  Coverage   99.03%   99.04%           
=======================================
  Files          46       46           
  Lines        4575     4585   +10     
=======================================
+ Hits         4531     4541   +10     
  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.