Read a transliterated Cyrillic initial opening a name as a given name, not a title (rules/H2, #1136) - #1294

Closes #1136.

rules/H2 (the leading-abbreviation title) read any unlisted word of two letters ending in a period as a title when it opened a name, so a transliterated Cyrillic initial became a title. Yu. A. Chibrikova read title 'Yu.', given 'A.', and Yu. Baryshnikov lost its given name. The same happened after a family comma (Chibrikova, Yu. A.) and behind a title (Dr. Yu. Petrov → title 'Dr. Yu.'), from 1.4.0 through 3.0.1. The Cyrillic original, Ю. А. Чибрикова, already read given 'Ю.'.

What changes. The shape now declines a closed set of transliteration digraphs, the Latin spellings of one Cyrillic letter, in any case: Yu, Ya, Ye, Yo, Zh, Kh, Shch and Dzh (_vocab._DIGRAPH_INITIALS). Both spellings of H2's test read the set: _vocab.is_title_shaped and the inline copy in _pieces.is_leading_title. The check is C-level and adds no frame.

Yu. A. Chibrikova    → given 'Yu.', middle 'A.', family 'Chibrikova'   (was title 'Yu.')
Chibrikova, Yu. A.   → given 'Yu.', middle 'A.'                        (was title 'Yu.')
Dr. Yu. Petrov       → title 'Dr.', given 'Yu.'                         (was title 'Dr. Yu.')
Ts. Dr. Norziha Che Him → title 'Ts. Dr.'   (unchanged)
Ch. Tahir Mehmood    → title 'Ch.'          (unchanged)

Which digraphs, and why this set. The set was chosen from three options (vocabulary word by word, a shape exception, the RU pack only) and narrowed to the digraphs that no corpus reads as a title. I counted every unlisted period word opening a name, or the part after a family comma, that master read as a title. The corpora were orcid-researchers, nepal-candidates, ctgov-officials, psc-individuals, opensanctions-parliaments, patentsview-inventors and jrc-names.

• Yu/Ya/Ye/Yo/Zh and Kh were never a title there.
• Ch reads about 8 times as an initial, against 4 uses as Chaudhry.
• Sh reads 2 times as an initial, against 4 uses as Sheikh or Shri.
• All 12 Ts. openers are the Malaysian professional title.

Those three keep the title reading, and what a Mongolian Ts. reads as is #1235's question. Th (Greek Θ, or Dutch Theodorus written short) is out by scope, not by evidence. The full table is in decisions/H2's 2026-11-09 #1136 entry.

What it reads as. The digraph is a name word, not the initial T5 tags. Over a grid of 24 shapes (the comma shapes, titles, a particle, a nickname, a suffix, a lone word) under all three orders, every digraph in the set reads field for field as J. does. The one difference is a lone digraph, which reports given-or-family where J. reports nothing. A caller who wants Yu. as a title still lists it as one (Lexicon.default().add("yu", kind="title")), which outranks the shape. The Library of Congress spellings (Iu., the tied I͡U.) are outside the set. The ISO 9 one-letter spellings were initials already, composed or decomposed.

Measured against master b89c548:

• The committed snapshot moves no name. The snapshot gains 11 rows, which are the new rule examples.
• snapshot.py --local --base <master> matches the base for nepal-candidates, ctgov-officials and patentsview-inventors.
• local_agreement.py --base <master> matches the base for psc-individuals, and changes 2 orcid-researchers records, both toward the boxes their filers typed.
• opensanctions-parliaments wasn't re-measured, because its cache predates the tool's dataset list. A scan of it found no digraph of the set opening a name.

Docs.

• rules/H2 gets a new Except bullet with examples, and its Signal line now names the digraph.
• decisions/H2 gets a Standing line and a Log entry. The entry records that the clause reads a closed word list, beside H2's split question.
• decisions/period-joined-words: its Should Yu. A. Chibrikova read Yu. as a given-name initial? Two-letter initials at the start of a name read as titles #1136 Open line becomes a Standing answer. The digraph set is a chunk the division reads on its own. Ts.Sandag-Ochir does not divide under the default vocabulary, and what it should read is left to Should a title glued to the given name by its period split? (Dr.John Smith gives title 'Dr.John', no given name) #1230 and Should an initial glued to a name split at its period? (Ц.Сандаг-Очир gives given 'Ц.Сандаг-Очир', no family) #1235.
• The root AGENTS.md gotcha on H2's shape now names the exclusion.
• docs/migrate.rst has a "Changed in 3.1" doctest, and docs/release_log.rst a bullet.

Reviews.

• A design-docs review of the first round found four issues: the Ts.Sandag-Ochir reading, the Th reason, a misplaced row and the signal. All four are fixed in 9637ffc.
• A review of that fix commit found three more: the caller's title override, Dutch Th., and ts listed under the vi language choice. Those are fixed in 7561f5b.
• A code review found no defects. Its suggested comma-count row is in.
• pr-test-analyzer and comment-analyzer: their findings are applied in the last commit (direct is_leading_title rows, a Vice Yu. Petrov example for the title-phrase head run, comment and release-log wording).

Checks run locally: the full suite on 3.11 and on 3.15 (in its own environment), the serial benchmarks (all 60 pass; earlier failures were machine load, and master passed under the same load), mypy, ruff, and both Sphinx builds.

Follow-ups, not in this PR:

• Md. as Muhammad written short is the largest population of this shape (2,785 ORCID credit names open with it), but it is Should a title word written after another title be the given name (Mr Sultan Khan, Mr Md Karim Uddin)? #853's question.
• The trailing Undram Ch. → suffix Ch. reported in the issue comment is untouched.

🤖 Generated with Claude Code