[derek73]

Phase 2k of #752: a given name written short outranks rules/H2's shape (H2 is the leading-abbreviation title). This settles the dotted half of #789. The bare half, wm out of the titles, was #873.

What changes

The dotted abbreviations below opened the given-name part as titles, from 1.4.0 through 2.6.0. H2's shape reads any unlisted word of two or more letters that ends in a period as a title there. Now each one is the given name it abbreviates:

input	before (1.4.0–2.6.0)	now
Chas. Smith	title Chas.	given Chas.
Mrs Ma. Liza Cruz	title Mrs Ma., given Liza	title Mrs, given Ma.
Smith, Ma. John (H2's own example)	title Ma.	given Ma.
SMITH, WM. DAVID	title WM.	given WM.
Chas. Smith, Ed	given Ed, family Chas. Smith	given Chas., suffix Ed (as John Smith, Ed)

• Vocabulary. Each word gets a given reading whose same_as is the full name, with its vetting counts:
  • en.toml: chas, thos, jas, geo, robt, benj, jno, saml, edw, richd, wm
  • a new fil.toml: ma (María) and jo (José)
• Build. same_as on a given reading names the full name. That name is not a vocabulary entry, so check 7 asks only that it be normalized and not the word itself. A family reading may not take one. The reading compiles into a new set, GIVEN_NAME_ABBREVIATIONS (Lexicon.given_name_abbreviations, re-exported from nameparser.config.given_names).
• Parser. Classify tags the word vocab:given-abbreviation. H2's shape test declines a tagged word, in both of its spellings: _pieces.is_leading_title reads the tag, and _vocab.is_title_shaped reads the lexicon. The second is also what C4's comma count and the title-phrase head run ask, which is why Chas. Smith, Ed moves. H2 itself has no special case.
• What stays put. A listed title still wins. The reading acts nowhere but the head of the given-name part, so these keep their readings:
  • the bare word: Ma Smith, Smith, John Ma, John Smith, JAS
  • the dotted word anywhere else: John Smith Ma. → family, John Ma. Smith → middle
  • ma's surname and credential readings
• Shim. The set is passed whole. A v1 caller who lists the word as a title gets 1.4's reading back.

Measured

All with --base against master 12384fa, 2026-10-26, and re-taken after the review fix with the same counts:

corpus	moved	shape
committed snapshot	2	Ma. opening the name, and H2's example
CLDR / GND / Wikidata agreement	0	the conventions view gains one Ma. title
ctgov-officials	15 of 536,981	each Ma. opening the name, mixed case
fec-individuals	24 of 1,893,964	all capitals, after a family comma: WM. 20, MA. 2, GEO. 1, EDW. 1
tse-donors	2 of 854,531	MA. opening the name
tse-donors spelling pairs	0
psc-individuals	86 fixed, 0 broken, 2 changed	all Ma.: 81 behind a salutation filed in the title box, 7 opening the name

These match the spike of 2026-10-24.

After merging master 32049ad (#896, phase 2j's wing), I re-measured against that master with --base and got the same counts in every row. Names that use both changes read as a plain given name would in their place (Mrs Ma. Wing Lau → title Mrs, given Ma., middle Wing).

Under a family-first order the word is now a name word placed by the order: Chas. Smith → family Chas., as Jack Smith reads.

test_a_given_name_written_short_reads_as_a_given_name checks this over 13 words × 6 spellings × 21 slots × 3 orders (4,914 texts):

• wherever the shape fired, the dotted spellings read as the name does with Jack in the word's place;
• everywhere, the bare spellings read as they do with the set emptied.

Each predicate half is mutation-checked:

switched off	pairs that fail
the peel (is_leading_title)	2,103
the comma count (is_title_shaped)	234
the title-phrase calls alone	117
the set put back in the wordlist rosters (pre-fix code)	66

Review

• Code review. It found that the new field had joined the two wordlist rosters in _vocab.py that only words claimed as non-names belong in: in_any_wordlist and claimed_as_non_name. That reversed John Smith, JAS and Jas SMITH, XYZ. The field now stays out of both, as bound_given_names does. AGENTS.md records the lesson.
• Second review of the fix commits. It found no regression: 0 bare moves against master over a 4,056-text grid. It did find one reading the H2 bullet had overclaimed. rules/C6's case contrast skips a period-marked word, so Chas. SMITH, XYZ reverses (given XYZ), just as J. SMITH, XYZ does today, while Charles SMITH, XYZ reads suffix XYZ. This is not new. I narrowed the sentence and left C6 alone; widening C6 would be its own decision.
• Design-docs review. It found stale "the shape outranks vocabulary" sentences in H2's Then, H4, H9, the H Background, decisions/A1 and mechanisms.md, and two README contradictions about same_as. All are fixed.

Axis 12, for your call. The H2 bullet reads no new signal in the name: the same shape, at the same position. Its one new input is the word's given reading, which outranks the shape. That is why it is H2's Precedence rather than a rule of its own. The Signal line says so.

Checks

• uv run pytest (also on 3.15), and -n0 -m serial
• mypy, ruff, build.py --check, sphinx doctest
• code review, design-docs review, and a second review of the two fix commits

Closes #789. Refs #752.

🤖 Generated with Claude Code