[derek73]

Phase 3e of #753: two vocabulary packs for #345's romanized Indic titles. Each is nothing but its choice of language, with no DEVIATES (the 3d pattern).

What ships

• locales.HI turns on smt, shrimati and shriyut (hi.toml, salutation readings default = false, after_given like their Devanagari spellings).
• locales.BN turns on smt, shrimati (bn.toml, the same readings) and adhyapak (academic, default = false).
• Example: parser_for(locales.HI).parse("Smt Nirmala Sitharaman") gives title Smt, given Nirmala. 1.4.0 through 2.8.0, and the default, give given Smt, middle Nirmala.
• Where a turned-on word reads as a title: wherever the title vocabulary is read: the head, behind a listed title, after a family comma, ahead of one (rules/H6), ending the given part (rules/H8), dotted at the end (rules/H5), and inside a period-joined token.
• No default parse moves. _compiled.py is current and the committed snapshot is unchanged.

Left out, measured

• shri. Its exclusion stays, with the counts.
  • ORCID's whole extract has 198 given boxes opening with Shri, nearly all given names (Shri Kant, Shri Lakshmi; 52 hold Shri alone).
  • Turned on, it broke 20 filed names and fixed none: PSC 2, ORCID 5, PatentsView 4, FEC 9 (BANGALEY, SHRI lost its given). The only fix anywhere was jrc-names' Shri Dharmendra Pradhan.
  • The clash is in the pack's own language, so neither prefer nor strict settles it. That is the RFC's same-language route (a given reading plus the title phrase, as default work), left Open.
• sushri. 14 of ORCID's 16 openers are Odia given names. It is left for strict, with no reading written: a plain hi choice would turn a default = false reading on for exactly the data that holds those names.
• The 3d question (a given reading of श्री in hi.toml/bn.toml): answered no. A Hindi or Bengali choice wants the salutation.

Moves under the choices

Measured 2026-10-30 on b57468d plus this change, with --base a detached master checkout.

• Committed corpora: none.
• CLDR locale view: hi+hi_Latn 31/48 and bn 14/24 both ways. The native-script hi-in (60/60) and bn-in/bn-bd (40/40) are unchanged too. Records gain the rows.
• nepal-candidates: 17 names under each choice (16 Shrimati, plus Smt.Nagiya, accepted).
• Pairs: 125,833 → 125,839 alike; 11 now alike, 5 now apart. All 5 pair the word with श्रीमति (short i), which hi.toml does not list (Open).
• Other local corpora: jrc-names, ctgov, psc, orcid, patentsview, fec and ukec show none under the final word set. That null is over populations holding no row with smt, shrimati, shriyut or adhyapak as a word; nepal-candidates' 17 rows are the only ones any source holds.
• Accepted costs:
  • Shrimati Kushwaha and Shrimati Shetty read with no given name.
  • NPPES's 4 SMT credentials read as the title in any comma part holding them (Jane Doe, SMT, Jane Doe, PhD SMT).
• shrimati ships, weighed: it leads as the title in all 16 nepal-candidates romanizations (by their Devanagari), against 1 ORCID person bearing it as a given name.
• shriyut and adhyapak ship with no measured leading use: no bearer of either opens a name (adhyapak's 17 bearers are all family names, written last). Both are admitted on their native spellings' title readings.

The full record and recipe are in the 2026-10-30 phase 3e entry of docs/design/decisions/vocabulary-model.md.

Also

• cldr_agreement.py maps hi, hi_Latn → HI and bn → BN, plus the native strata hi-in, bn-in and bn-bd.
• Rotators cover every changed word in each shape.
• Docs: docs/locales.rst (table, a "Romanized Indic titles" subsection) and the README/locales available() doctests.
• An Unreleased release-log bullet.
• A note in nameparser/vocab/AGENTS.md: count the boxes a title candidate opens, not only the boxes holding it.
• The second commit is the fix round from a design-docs review: the slots the titles reach (H5/H6), adhyapak's cost, the Recompute recipe, the RFC status line, Shreemati in the Open.
• The third commit restates the evidence after the two PR reviews:
  • the CLDR basis of the श्री answer (bn unmeasured);
  • the dotted forms' moves named by site;
  • the two no-given moves, the shrimati weighing and the population behind each null;
  • the NPPES SMT cost;
  • hi/bn added to the contributor checklist.
• The fourth commit is the second-round fixes:
  • the Standing criterion restated for words with no measured leading use;
  • the dotted forms' sites taken from a comma-shape sweep (rules/C2 included);
  • SMT's cost covering every comma part that holds it;
  • both sides of Srimati's count.

Checks

• uv run pytest, -n0 -m serial, mypy, ruff, sphinx-build -b doctest and python -m doctest README.rst: all pass.
• The suite on 3.15 (a separate venv) and without namedivider: pass.
• test_locales and test_cldr_agreement with namedivider: pass.

🤖 Generated with Claude Code