[derek73]

Closes #1014.

What changes

capitalized() no longer gives an inner capital after a Mac prefix by shape alone:

input	before (1.4.0 – 2.8.0)	after
JOHN MACHADO	John MacHado	John Machado
MARIA MACIAS	Maria MacIas	Maria Macias
JAMES MACKENZIE	James MacKenzie	James Mackenzie
RONALD MACDONALD	Ronald MacDonald	Ronald MacDonald (mask)
J.MACDONALD SMITH	J.MacDonald Smith	J.MacDonald Smith (chunk mask)
JOHN MCDONALD	John McDonald	unchanged

• Mac is title-cased unless the vocabulary records its casing. 42 Gaelic surnames written MacX by most of their bearers (more than half of at least three writings) carry a cased mask on a family reading. Name readings may now carry cased (_reading_index.BEHAVIORAL), and an entry holding only such a reading compiles, so the build no longer warns it inert.
• Where the masks live. They are in en.toml, the anglicized spellings being English ones. They are listed again in gd.toml (37) and ga.toml (the Irish MacCarthy, MacMahon, MacManus, MacDermott), so parser_for(locales.GD/GA, strict=True) keeps them.
• A chunk after a full stop takes its own mask, as a hyphen part already did. This keeps J.MacDonald, and it lets credential chunks take theirs in forced repair (Phd.Md → PhD.Md, Msc.Ed. → MSc.Ed.).
• Mc keeps its convention. _render._MAC becomes _MC, matching Mc alone (nameparser.config.regexes, which held the public copy, left master in Remove nameparser.config.regexes in 3.0? The parser has not read REGEXES since 2.0 #1008).

Measured (2026-11-02)

The sources are ORCID's family-name box plus the names in the UK Electoral Commission donors, GOV.UK people and ClinicalTrials.gov officials local corpora. Each writing was counted as typed; the recipe is in decisions/R4's 2026-11-02 #1014 entry.

	writings	match the writer
Mac: old rule (always MacX)	4,616 of 970 words	16.3%
Mac: plain title case		83.7%
Mac: title case + 42 masks		91.2% (in-sample)
Mc convention	6,312	96.8%

Bearers of the Gaelic names split by family: MacDonald 143 vs Macdonald 77, Mackenzie 59 vs MacKenzie 44. So no shape can decide them, and a per-word mask can.

Accepted costs:

• An unlisted Gaelic name loses its capital: SHIRLEY MACLAINE → Shirley Maclaine. The release log points callers to Lexicon.capitalization_exceptions.
• Mc names that aren't Gaelic stay McX: Mchunu → McHunu.

Moves

• Committed snapshot: case views only, no role moves. Czech Macáková/Machová, Macías and Machbuba are fixed; Shirley Maclaine and the fictional macdole-eisenhower are the accepted cost; forced Msc.Ed. → MSc.Ed..
• snapshot.py --local ctgov-officials --base master:
  • 619 moves of names containing mac.
  • 126 of names without it, all credential chunks.
• ukec-individuals: 69 moves, all Mac names.

Docs

• rules/R4 (case repair reads the unjoined mark): a new bullet on Mac masks, a full-stop example, and the Mc wording.
• rules/R5 (the case-repair gate): its examples are renamed from Shirley Maclaine to Ronald Macdonald, since the forced repair of the old one now changes nothing.
• decisions/R4 and R5: Standing bullets and log entries.
• Also updated: the vocab README, the release log and rules_budget.json.

Checks

• Full suite, -m serial benchmarks, mypy, ruff and build.py --check pass.
• New tests: a mask-only surname entry compiles and isn't inert, with a negative control on each half of the model change. Strict GD/GA packs keep the masks, with a negative control: the strict rows fail without the gd/ga entries.
• The design-docs reviewer ran two rounds. The first found six issues, all fixed: chunk masks, strict-pack claim, Mc wording, recipe numbers, Mc recipe, stale Standing line. The second found the fixes sound and two notes, both recorded in the decision entry.
• Not run: the suite on Python 3.15 locally. CI covers it.

🤖 Generated with Claude Code