[derek73]

Sub-phase 3h of #753's work order: the Irish, Scottish Gaelic and Brazilian Portuguese packs. Each is a Locale that is only a choice of languages (no lexicon fragment, no policy patch), so it declares no DEVIATES: its surface is the words its choice changes (_reading_index.changed_words), which the non-interference gate derives. The default does not move.

What each pack turns on

Pack	Chooses	Readings turned on (default = false in the file)
locales.GA	ga	clan nic (daughter of), mhic and uí (married forms of mac, ó), mic (genitive of mac)
locales.GD	gd	clan nic
locales.PT_BR	pt-BR (and pt)	agnomen sb (Sobrinho), cased = "SB"

nic and mic carry given readings (US SSA 77 and 26, Wikidata 173 and 10), so a leading Nic/Mic stays the given name in a name read given-first, reported PARTICLE_OR_GIVEN (rules/P1), as Mac does by default. nic, mhic, mic and uí carry cased masks, so case repair keeps their capital under the pack. nic's and sb's exclusions become these readings (item 9 of the 2026-10-27 #753 entry).

parse("Máire Uí Bhriain")                       middle 'Uí', family 'Bhriain'
parser_for(locales.GA).parse(...)               family 'Uí Bhriain'
parse("JOSE CARLOS SILVA SB")                   family 'SB'
parser_for(locales.PT_BR).parse(...)            family 'SILVA', suffix 'SB' (reported)

Departures from the brief, for review

• F and N are not turned on: they stay excluded, the pack included. Turned on and measured on tse-donors (the corpus they were proposed from), scoring each moved name by what the registry wrote at the moved word: F was the agnomen in 48 moved filed/registry pairs and a letter abbreviating a surname in 104 (ANA MARIA F / ANA MARIA FILGUEIRAS), N 18 vs 59 (EDINEI VITAL N / ... DO NASCIMENTO). The strict "pair reads alike" measure (given, middle, family, suffix each equal) had shown 34 fixed and 0 broken for the three letters together, because most moves went from one wrong reading to another; a note in mechanisms.md#PAIRED-SPELLINGS-PER-CHANGE records the trap. Reverting is two entries in pt-BR.toml if the call goes the other way.
• mhic is in ga.toml only. In Scottish Gaelic it mostly links a patronymic chain (Iain mac Dhòmhnaill mhic Sheumais), which listing it splits at its last link, as the default already splits … mac … mac …. The same split happens under ga for Seán mac Seáin mhic Thomáis (no corpus name moves), accepted for the married form Máire Mhic Gabhann.

Measured (2026-10-30, master 4067514 + this change, Python 3.11)

• Default: _compiled.py unchanged; the committed snapshot gains only the three names rules/S11's new examples add to the rules corpus, no entry moving; snapshot.py --local tse-donors --base <master> matches the base.
• Committed corpora (snapshot.py --languages): ga 4, gd 4 entries, all mc: the choice prefers ga's/gd's clan mc over en's Military Cross credential (John Smith MC → family MC), a 3b prefer effect on the mc both files already held. pt-BR moves 40: pt's 38 plus two of rules/S11's new SB examples.
• CLDR locale view: ga 14/24, gd 15/24, pt (CLDR's Brazilian file) 17/24 with and without the pack; new rows recorded in cldr_agreement_locales.json.
• tse-donors, pt-BR against pt: 14 more names move; 13 are SB read as the agnomen, the registry spelling SOBRINHO in every one (pairs reading alike 11 → 22); 1 is the SB case mask alone.
• PSC, every record holding a changed word: ga 4 fixed (Clodagh Nic Lochlainn, Orlaith Mhic Mhathuna, Damien Patrick Mc) / 1 broken (Nicholas Nic Rawlings Limited Rawlings, a company filed as a person); gd 3 / the same 1.
• ORCID, every country: ga 24 fixed / 0 broken (Pauline Uí Dhuibhir, Éimí Mhic an Ridire, 22 Nic …); gd 20 / 0.
• ctgov-officials, pt-BR against pt: 17 more; 12 only capitalized(force=True) (a mid-name SB keeps capitals), 5 where the S.B. degree now fills generation instead of credentials (Howard S. Gordon, MD SB), one of them moving fields (Silvio Brugger, SB: given SB → suffix). Recorded as an accepted cost of choosing Brazilian Portuguese.
• pt-BR's moves in PSC (12 broken) and ORCID (21 fixed, 11 broken) are all pt's own: see Open.

Open

• For 3g: choosing pt (so pt-BR) prefers pt.toml's title sr (Senhor) over English's generation Sr: Gary Holloway Sr → family Sr, J. Smith Sr. → title Sr. (PSC's 12). Recorded in the decisions entry.
• tests/v2/test_locales.py's _is_vocabulary_pack/_deviates helpers are also being written by the 3d (ne) branch; whichever lands second keeps one copy.

Same-PR docs

rules/S11 (SB under the choice, with [pt_br] examples and F as a boundary), decisions/vocabulary-model (Standing line + 2026-10-30 phase 3h entry), decisions/S11 (Standing line amended, Log entry), mechanisms.md (field note above), RFC status and work-order row, docs/locales.rst (table rows, a "Packs that choose a language" recipe doctest, contract steps for a vocabulary pack), README's available() doctest, release-log bullet.

Review

A design-docs review pass found four figure problems (an unregenerated rules corpus behind a "no re-record" claim, the S.B. degree's cost unmeasured, two pair figures without their comparator, and 242 counting pairs not distinct names); all fixed in the second commit.

A second review (orchestrator) found the missing Irish case masks (MÁIRE NIC GIOLLA PHÁDRAIG repaired to Máire nic under ga/gd), docs presenting the MC/Sr readings as features, an unqualified leading-Nic claim, the SB mask's reach, two wording slips, and a missing step in the F/N recipe; all fixed in the third commit, with failing-first rows for capitalized() under each pack.

Checks

uv run pytest, -n0 -m serial, --python 3.15, mypy, ruff, build.py --check, sphinx doctest + html (clean), python -m doctest README.rst. namedivider is not installed locally, so the ja rows ran skipped, as in CI's build jobs.

🤖 Generated with Claude Code