[derek73]

What

Follows the mul.toml audit decided in chat on 2026-11-03 ("ms, miss, mba"), measured with the strict-gap report of #1025.

• ms and miss move from en.toml to mul.toml, each whole. ms moves with its credential and miss with its family reading (check 13 of tools/vocab/build.py refuses a mul.toml word whose on-by-default name reading stays in another file; the family reading derives no gate on a salutation). Every strict choice reads mul.toml (Should a strict choice still read PhD, Dr and Prof? (parser_for(locales.PT, strict=True).parse("Ana Souza, PhD") → given 'PhD') #1005), so a strict choice now reads both salutations as mr and mrs already were.
  • No default parse moves for either, and no plain choice does (en.toml was the one file holding a reading of either).
• mba gains a family reading. It is the Igbo surname Mba (US Census 2010: 320 bearers) and a Fang name part.
  • This supersedes the 2026-10-21 hold-out in decisions/suffix-acronym-collisions, which kept it a plain credential.
  • Chinedu Mba reads family 'Mba', as 1.4.0 read it, where 2.0.0 through 2.8.1 read suffix 'Mba' and no family.
• The rest of mul.toml stays, ing included. The audit judged every word against the Should a strict choice still read PhD, Dr and Prof? (parser_for(locales.PT, strict=True).parse("Ana Souza, PhD") → given 'PhD') #1005 criterion; the evidence is in decisions/vocabulary-model's new 2026-11-03 entry.

Measured

On 7ce9cbf with config_diff.py at 22234c1, the first 20,000 records a group.

• Strict losses (gaps, 83 tags). A broken record is laid to the word whose own token moved.
  • ms broke 38,117 records across tags and fixed 11; miss broke 28,537 and fixed 2. Nearly all are the UK register's title boxes.
  • On this branch both are 0. Strict zh falls from 4,832 broken to 965, fil from 4,899 to 1,066, pa from 3,008 to 155, ro from 2,577 to 60 and pl from 2,093 to 70. Strict en stays at 7,362.
• Why ms moves whole. With its salutation alone in mul.toml, strict ko, hi, pl and zh read MS past a comma as a title: 46,838 of 51,461 names moved (A Israni, MD, MS → title 'MS').
  • Moving whole costs the South Indian initials Should a strict choice still read PhD, Dr and Prof? (parser_for(locales.PT, strict=True).parse("Ana Souza, PhD") → given 'PhD') #1005 kept it for. Under strict without en, Radha MS reads suffix 'MS', as the default does. About 150 ORCID and PatentsView India records hold MS in the family box.
• ing. Moving its credential to cs, sk, de, ro, nl, fr, sr and hr and its names to en and sv was measured under all 162 configurations. It was worse both ways:
  • those eight languages' strict choices lose the gate (Alvin Ing → suffix 'Ing');
  • every other strict choice loses the credential (Daniela Ponce, Ing → given 'Ing').
• mba, default against the base (measure.py --base, 22 steps, none failed). Every moved name in every local corpus holds mba: ctgov 20 entries, FEC 179, ORCID 26 against 7ce9cbf (27 against the merged master, by the review), PSC 12, JRC 12, GOV.UK 1.
  • Of the 58 names whose fields move, about 49 read the name better (Léon Mba, Patrick Arthur MBA ABESSO).
  • About 9 read a degree into the name, the cost the class pays for ma and meng. Where a name word can stand, the degree reads as the name: GULLEY MBA, KELLEY D. → family 'GULLEY MBA', Smith, MBA Jane, , MBA and RUH MD,MBA, RICHARD → given 'MBA'. A period-marked Smith, Mba. reads title 'Mba.', as Meng. does; brackets configured for maiden names read (MBA) as the maiden name, as (MA) and (MEng) do.
  • John Smith, MBA, John Smith MBA (in capitals) and Andrew Perkins (MBA) are unmoved.
• Committed corpora. The snapshot moves john smith mba (ambiguities only) and the two rule examples below. CLDR, GND, Wikidata and conventions agreement are unchanged.

Docs in the same diff

• decisions/vocabulary-model: the audit entry, its Standing line, and the Should a strict choice still read PhD, Dr and Prof? (parser_for(locales.PT, strict=True).parse("Ana Souza, PhD") → given 'PhD') #1005 entry's ms clause marked superseded.
• decisions/suffix-acronym-collisions: the mba entry and Standing; the 2026-10-21 hold-out marked superseded.
• Two rule examples that used MBA as the unambiguous credential take another:
  • rules/S1's (PhD MBA) becomes (PhD MSc), with decisions/S1's Standing line and the 2026-10-21 entry's head marked;
  • rules/M1's Andrew Perkins (MBA) under maiden-parens becomes (PhD). A marked member in a maiden pair reads maiden, as (MEng) does.
• nameparser/vocab/AGENTS.md: a dual-reading word moves whole, with the ms measurement.
• vocab/README.md's mul paragraph, locales.rst's list of shared words, the RFC's strict-losses note.
• A release-log bullet for mba.

The rules corpus is rebuilt and the snapshot re-recorded.

Review round (5310083)

An independent review found no behavior defect; it fixed the prose and the stated limits: the 2.8.1 range, the release log's limit sentence and the decisions entry's costs (the names above), stale Ms and (MBA) examples in locales.rst, usage.rst and AmbiguityKind.SUFFIX_OR_NICKNAME's comment, the RFC's supersession note, miss's reason, and vocab/AGENTS.md's count of checks, with a note that check 13 already refuses splitting a word.

Checks

• uv run pytest -q: 35,238 passed. -n0 -m serial: 42 passed.
• uv run mypy, uv run ruff check.
• sphinx-build -b doctest and -W -b html; python -m doctest README.rst.
• tools/vocab/build.py --check.
• Not run: Python 3.15. No code changed, only vocabulary data, tests and docs.

The audit's measurements use tools/differential/config_diff.py from #1025; this PR does not depend on it to build or test.

🤖 Generated with Claude Code