[derek73]

Sub-phase 3d of #753's work order: the first language pack, and the pattern 3e–3h copy where a pack is a choice alone.

What changes

• nameparser/vocab/ne.toml gains given = { default = false, note = … } readings on श्री, guru and गुरु. They go in ne.toml because a chosen tag selects its own file and its ancestors' only (3b), and ne has none. The title readings stay where they were. A name reading never prefers, so under ne each word holds both readings and derives TITLES_AMBIGUOUS, a title only inside a title phrase (rules/H9, the title phrase).
• locales.NE = Locale(code="ne", lexicon=Lexicon.empty(), languages=frozenset({"ne"})), registered in locales/__init__.py. It declares no DEVIATES.
  • tests/v2/test_locales.py's _deviates(code) ors the surface derived from _reading_index.changed_words into the predicate of every pack with languages. A hand DEVIATES therefore needs to declare only what a fragment or patch beside a choice changes. Nothing refuses one that restates the choice's words.
  • The registry contract refuses any DEVIATES attribute (even None) on a pack that is nothing but a choice, and requires one of every other pack. The rotators cover each word in each script, at the head, behind a listed title and after a family comma.
• local_corpora.py alike --languages: the pair measure under a choice, against the default's flags (the tree's own or --base's), as snapshot.py --languages does. It re-keys every kept alike base once (_parse joins _ALIKE_CODE).
  • Both flags read the bare choice, not a registered pack's fragment or patch. That is the whole of NE; both tools' docs state the limit.
• cldr_agreement.py maps CLDR's ne file and Wikidata's ne-np stratum to NE. cldr_agreement_locales.json gains NE's rows, and neither score moves.
• hi.toml: not touched. Its श्री note (a given reading there would make the title one only inside a title phrase) stays true of the default. 3e can edit it freely.

The default does not move

build.py --check passes, _compiled.py is unchanged, and _index.py gains three rows. tests/v2/test_snapshot.py passes with no re-record.

Under a family-first order

The words are name words placed by position. Under a family-first order they land in the family name: श्री राम बहादुर थापा gives family श्री, with given राम under FAMILY_FIRST and given थापा (middle राम बहादुर) under FAMILY_FIRST_GIVEN_LAST. The docs and the release bullet say so.

What ne moves (2026-10-30, on 40675146 plus this change)

• Committed corpora (snapshot.py --languages ne): 6 entries.
  • 4 are CLDR's hi, kok and mr renderings opening with the salutation श्री (CLDR title श्री). They lose the title, the cost the user docs state: use the pack for Nepali data.
  • Guru Gobind Singh and Guru Nanak read given Guru.
• CLDR per pack: ne stays 15/24 and Wikidata ne-np 10/10, both ways. Neither file holds one of the words.
• nepal-candidates (snapshot.py --local nepal-candidates --base <checkout> --languages ne): 296 names move. In every one the word leaves the title for the given name: 209 श्री, 42 Guru, 41 गुरु, 1 GURU, 1 श्री., 2 सु.श्री.
• Pairs (local_corpora.py alike nepal-candidates --languages ne): 125,833 → 126,040 of 126,981 read alike, with 211 now alike and 4 now apart.
  • All 211 hold श्री, which the romanization writes as a name word (Shree 191, Shri 6, Sri 2).
  • Of the 4 now apart, 3 romanize श्री as Mr.. The fourth is the open defect below.

Open

• A glued period-joined title reads as a name word under ne, in both directions. period_joined_vocab makes a token a title when ANY chunk is in titles, and श्री has moved to titles_ambiguous.
  • It costs सु.श्री. (Sushri, written with its periods): 2 nepal-candidates names read given.
  • It fixes श्री.राम थापा, which reads given श्री.राम instead of title.
  • A fix would move the default for every TITLES_AMBIGUOUS word, and its grid should hold both shapes.

Recorded in decisions/vocabulary-model's 2026-10-30 phase 3d entry and decisions/deviates-registry. The RFC status line is updated.

Checks

• uv run pytest: 33,000 passed, on 3.11 and on 3.15 (-n 4; the default -n auto hit an xdist INTERNALERROR at load ~100).
• mypy, ruff, build.py --check, sphinx-build -b doctest and python -m doctest README.rst are clean.
• pytest -n0 -m serial: 42 passed at load ~30. Earlier runs at load 50 to 100 failed a different set of wall-clock ratios each time, and so did master at 40675146.
• The ja extra was absent for every run.

🤖 Generated with Claude Code