[derek73]

Closes #345 and #870.

These are the romanized Indic honorifics left over after phase 3e (locales.HI, locales.BN), plus the default reading of Devanagari श्री. Each value was decided in chat on the counts below.

What changes

• Six words are excluded with their counts and read as a title under no pack: pandit, sardar, kumari, babu, acharya (hi.toml) and begum (bn.toml).

  • Each opens far more names as a given name or a surname than as the title.

  • ORCID's whole 2025 extract, counting given boxes that open with the word:

    Word	Given boxes
    kumari	520
    babu	466
    sardar	330
    begum	236 (Turkey 69: Begüm)
    pandit	33
    acharya	24

  • Family boxes add kumari 5,059, acharya 2,068 and pandit 951.

  • nepal-candidates, PSC and FEC point the same way.

  • Title uses found are few: jrc-names news text has at least 6 Begum and 10 Sardar (Begum Khaleda Zia, Sardar Patel).

• sushri is now read as the title only under a strict Hindi choice.
  • hi.toml gains a salutation reading, off by default.
  • or.toml declares the Odia given name (Sushri Sangita).
  • So parser_for(locales.HI) treats it as a title only inside a title phrase (rules/H9). parser_for(locales.HI, strict=True) reads Sushri Sangita Behera as title Sushri.
  • strict_ungated's reviewed table gains hi: sushri.
• The romanized janab is dropped. Three records are no evidence either way. ur.toml's "unverified" exclusion is removed rather than kept as a refusal, and Should romanized Janab read as a title, as Devanagari जनाब and Bengali জনাব do? (Janab Ali Sai reads given Janab) #1043 is closed. The native जनाब and জনাব stay salutations, since CLDR's bn renderings state জনাব as the title and no source counts either as a name.
• श्री stays the default salutation (Should Devanagari श्री read as a name word when the name's romanization keeps it? (श्री कृष्ण श्रेष्ठ reads title श्री, Shree Krishna Shrestha given Shree) #870).
  • Hindi, Marathi and Konkani write it as Mr. Nepali writes it as a given name's first word: nepal-candidates romanizes it as a name word in 202 of 216 names.
  • locales.NE already reads it that way (phase 3d).
  • rules/H9 gains an Accepted bullet with a [ne] example recording the cost.

What moves

• The default and every plain choice: nothing. _compiled.py is unchanged, and the only committed-corpus change is H9's new example joining the rules corpus.
• Strict HI: 4 of 12,952 local names, all sushri read as the title, measured with config_diff.py diff over every local corpus with --readings in both trees.
  • Three are right: two nepal-candidates names written सुश्री, and Sushri Umashri Bharti.
  • One is the Odia-name cost: Sushri Das Bristi.

The recipes are in decisions/indic-honorifics' "2026-11-03 (Derek), #345/#870" entry.

Docs and tests

• docs/locales.rst gets a strict Sushri doctest, the list of words no pack reads, and drops the stale "Ship hi and bn locale packs for Latin-transliterated Indic honorifics #345 in progress" pointer.
• The release log's HI bullet is amended.
• The RFC's 3e row and overview, and vocabulary-model's 3e Standing line and log, are updated.
• tests/v2/test_locales.py gains HI pack contents, a rotator and a strict test.
• rules_budget.json is re-recorded: H9 grows by one Accepted bullet that reads H9's own signal, a title word that may be a given name.

Review

• Independent code review and design-docs review.
• A second-round review of the fix commits.
• Every finding was applied. The review moved श्री's cost from H1 to H9, where its signal belongs.

🤖 Generated with Claude Code