[derek73]

Follow-up to #1146 and #1147: the "native-script vocabulary" wording sweep #1147 listed, plus the findings of a post-merge review of its last fix commit (af7431c).

What changes

• A native-script word under a strict choice. strict_kept keeps a reading when its file is chosen, it is mul.toml's, or its word has a letter and no Latin letter, and choose then runs as under a plain choice. So a strict choice keeps a native-script word's readings, name readings included, exactly as a plain choice of the same languages keeps them: a default = false reading stays off unless its file is chosen. Strict pt reads गुरु प्रसाद शर्मा title गुरु, ne.toml's given reading of it being off; strict ne reads given गुरु.
  • Seven sites said "native-script vocabulary", leaving the name readings unsaid: Lexicon.strict's docstring, parser_for's docstring and its no-language refusal, the UNCHOOSABLE comment, root AGENTS.md, mechanisms.md, decisions/vocabulary-model's Standing summary, and the vocabulary-model RFC. Dated decisions entries and the RFC's phasing row are left as written, as history.
  • Bring the user docs in line with 3.0's readings, views and language choices #1146/Add a Speed and testing section to the README #1147's docs/customize.rst and docs/AGENTS.md said a word in another script keeps "every file's" readings, which the default = false case refutes.
• README, building a parser. The first build of a choice loads the reading index only unless a parse has already loaded it: after a default parse of a name holding a listed word (Dr. John Smith Jr.), the first parser_for(locales.ES) took about 12 ms, not 37–43. The figures stand; the sentence now says the index is "about two thirds of the cost", "unless a parse has already loaded it".

How it was checked

• On this branch: for every one of the 243 native-script words in the reading index, readings_of under a strict choice equals the plain choice's, for pt, ne, hi, ru, ja, ko and en (0 mismatches). Strict en keeps uk.toml's family reading of пан.
• ruff, python -m doctest README.rst, sphinx-build -b doctest and -b html (no output), and the doc-citation, design-docs, rules and reading-index tests pass.

Reviews

• pr-review-toolkit:comment-analyzer, post-merge, on af7431c: 13 claims exact, 2 overbroad (the README first-build scope; "every file's" for a word in another script). Both fixed in 03f10ed, the first commit here, which also extended the qualifier to the sweep's own wording: a default = false reading stays off.
• pr-review-toolkit:comment-analyzer on 03f10ed: the claim holds over every tag alone and in pairs (0 differences). Two RFC sentences (verb agreement, the table row), the README's "compiles" for the index (it loads it), and RFC sites at the strict hu/ko sentences and the no-language refusal. Fixed in b7d8c11.
• design-docs-reviewer on 03f10ed: the RFC overview still had the old definition, and the Standing line said what no dated entry said. b7d8c11 added a dated 2026-11-07 Say what a strict choice keeps of a native-script word #1148 entry and the overview fix.
• second round, design-docs-reviewer on b7d8c11: "clarified by" is not an edit decisions/README.md allows, so the entry is now a partial supersession; it named strict_kept for 3j, when the filter was then strict_pool's; the RFC's Status paragraph and 3j row, and strict hu read as the default's. Fixed in d83c3c8.
• third round, design-docs-reviewer on d83c3c8: no confirmed defects (strict hu equals the default's 293 native-script readings; quote verbatim; 3,240 choices, 0 differ). One low finding, the 3j head lacking the Say what a strict choice keeps of a native-script word #1148 note, fixed in the last commit, a one-clause addition checked by the doc tests.

🤖 Generated with Claude Code