[derek73]

Part of #761.

This PR moves H1–H6 (the titles and honorifics section) from the block format to the skeleton format, one commit per rule. Later commits fix review findings and carry out Derek's two decisions (below): H2 states that C13 wins, and H4's join clause becomes H7. No parse changes, and the migration commits change no rule's meaning:

• Every example line is kept verbatim. Each rule's example lines match master as a multiset, and H2's two NFD-typed lines were copied byte for byte from git show.
• The migration commits move neither the corpus nor the snapshot. The decision commits only ADD names, for their new examples.
• All 18 code-comment quotes of H1–H6 still match verbatim, so no code comment needed retargeting.

tests/v2/rules_budget.json is re-recorded. Each H row changes format, showing as its old row beside its new one. The word counts grow because bullet leads were added; no new clause reads a new signal.

Where the old clauses went

H1

• Then: the old statement's three code-quoted clauses. The behind-run and comma clauses each became a Then bullet.
• Except: "A particle surname a comma or P1's fold settled keeps the family", taken from the old Accepted clause. Scoped after review: with no comma, Sir van Gogh reads given 'van'.
• Accepted: the clauses on the empty family, the listing record (with the 1.4.0 note), unknown title words and lord/lady (Should Prince Harry read Harry as a given name? (prince, princess, lord, lady are titles but not given-name titles) #519), and graf/greve/herra.

H2

• Then: the shape rule, with "the shape outranking vocabulary where it fires" written in.
• Then bullets: the bare initial (now pointing to T5), and NFD (A decomposed (NFD) name loses an abbreviated title: Dr. inž. Janez Novak reads inž. as the given name #652).
• Except: the family-comma credential run (C8). This merges the statement's exception with its Accepted clause.
• Accepted: the abugida clause, the CJK veto, and the before-family-comma clause (H6).

H3

• Then: the chain rule.
• Then bullet: the floor (giving the last word back).
• Accepted: the three Accepted clauses, unchanged.

H4

• Then: the two halves. A single title word is a Then bullet, as is the old "ONE site … declared order" clause (a reports clause).
• Except: "a title in front takes the input out of the suffix half" (MD DDS).
• Precedence: H2 over a dotted lone credential (Jr.).
• Accepted: the CJK-honorific scope and the join reach.

H5

• Then bullets: transparency, input order, a title among post-nominals, and what reports. These were Accepted clauses that state results.
• Precedence: a particle chain before a trailing title, the family-comma particle chain, a maiden clause, and the bound given-name word (P5). These were the two long Accepted clauses, split.
• Accepted: the ordinary-surname reach (Mary Jane King.).
• "The chain" named explicitly: "the title chain" where the old paragraph meant H5's own chain. Where it meant the given part's particle chain (the rules/H5 comment in _group.py), it now says "particle chain".

H6

• Then: the run and its order-independence.
• Then bullet: vocabulary only, never H2's shape (Sta. Maria).
• Except: the particle stop, the post-nominal stop (H3), and leaving the family a word.
• Accepted: ordinary-surname titles, and the German ranks. "The other German ranks" became "German ranks other than 'Freiherr'", since the sentence lost its referent when the particle-stop sentences moved out. Titles ∩ particles is exactly {freiherr, st}, so the meaning is the same.

Old interacts: IDs

H1 (old: C3 C8 O5 H3 H5 H6 P2 P3 P5 M2 S1 S2 N1 N3)

• Kept: C3, O5, H5, H6.
• Added: P1. The old text already named "P1's fold".
• Dropped as noise:
  • S1, S2, N1, N3, M2: the clause "whatever suffix, nickname or maiden name stands beside it" names these by meaning.
  • P2, P3, P5: what counts as one name word. The examples van der Berg, Sir and Sheik Abu Bakar carry it.
  • H3: the run is "a run of titles".
  • C8: the comma bullet states the no-given-part case in words.

H2 (old: C5 C6 C8 S3 P4 H5 H6 W3 W4 S9 T5)

• Kept: all except P4, H5 and W4.
• Dropped:
  • P4: no H2 clause reads the particle-opening position.
  • H5: the front/back contrast lives in the H Background.
  • W4: the CJK veto is a script-table question, and its reading is W3's, which stays named.

H3 (old: C1 H1 H4 H5 H6 S2)

• Kept: H1, H4, H6.
• Dropped:
  • C1: "before a family comma" says it.
  • H5: no H3 clause reads the trailing run.
  • S2: "suffix vocabulary"/"post-nominal" say it.

H4 (old: H1 H2 H3 H5 S2 O5)

• Kept: H1, H2, H3.
• Added: W2, P3, P4. The old text already named them.
• Dropped:
  • H5: H5 points at H4's report, not the reverse.
  • S2: post-nominal vocabulary is named in words.
  • O5: the order-independence bullet says it in words.

H5 (old: H1 H2 H3 H4 M2 P2 P5 P6 S2 C1)

• Kept: all except H3 and C1.
• Added: P3. The old text already named it.
• Dropped:
  • H3: "any leading run" says it.
  • C1: "after a family comma" says it.

H6 (old: C1 C3 H1 H2 H3 M2 P1 P2 P4 S2 W3)

• Kept: C3, H1, H2, H3.
• Dropped:
  • C1: "family comma".
  • P1, P2, P4: the particle stop says these in words, with examples Dr. de la Vega, Juan and Freiherr von Berg, Ed.
  • S2: the post-nominal stop goes through H3.
  • M2: no H6 clause reads a maiden name.
  • W3: no H6 clause reads CJK.

Decided (Derek, 2026-10-20): the two gaps this migration found

His words: "Yes to both: C13 wins over H2 for a listed suffix word (Precedence bullet on H2, no behavior change), and H4's join clause moves to its own new H rule. Do it in PR #779."

1. C13 wins over H2 (cf93669). H2 gains a Precedence bullet, and its Then names the exception. New examples: Smith, Esq. John → suffix "Esq. John", Smith, Insp. John → title "Insp.", Smith, Ma. John → title "Ma.". decisions/H2 has the dated entry.
   • Scope confirmed by Derek 2026-10-20: credentials that start S8's run, no behavior change. "A listed suffix word" and "no behavior change" cannot both hold. I measured every alphabetic suffix word of two or more letters in Smith, <Word>. John:
     • 666 read as the postnominal part.
     • 22 still read as a title by H2's shape: the 12 title/suffix duals (Cpl., Ms., Sr.…), 7 ambiguous acronyms (Ma., Do.…), and Abd., Mc. and Vd..
     • S8 bars all 22 from starting its run.
   • To keep the parser unchanged, I scoped the bullet to C13's own words, "a suffix word that starts S8's run", and pinned Smith, Ma. John beside it. Extending the win to all 22 would be a behavior change, and is not proposed.
2. H7 — a joined name unit carrying title vocabulary (4be2b4a), split out of H4's join clause.
   • Text: verbatim save two edits, which decisions/H7 records: "as well" dropped, and the corpus count dated ("reached on 2026-09-08").
   • New: a When, a Because, and a bullet pointing to H4 for a unit of one word. H4 gains a Then bullet pointing to H7.
   • Pointers: O5's pointer and its interacts: name H7. _assign.py's emitter comment quotes H7, so H7's implemented: holds. Two cases.py notes now name H7.
   • Examples: the old clause had none (its inputs were pinned in cases.py), so H7's are new. Prince of Wales is the boundary.

Corpus and snapshot: corpus_rules.jsonl and snapshot.jsonl gain the new examples' names (3 for H2, and 6 more for H7, with Attorney General of Minnesota already in another corpus file). No existing reading moves (snapshot.py --record added names only).

Review fixes

• a21523a (second review round):
  • H1's Except holds only the rows where the family is kept, and the ambiguous/bare sentence returns to its Accepted bullet.
  • H6's Sta. Maria row (outside When) moves to the main fence, and the German-ranks clause now reads "German ranks the title vocabulary lists". The old wording overclaimed: Markgraf, Landgraf and Freiin are unlisted and do not peel.
  • H2 now says "single-letter initial", so its T5 pointer reads in T5's terms.
  • H5's bound-word lead no longer contradicts its body.
• a839d5a: O5's interacts: gains H7, and decisions/H7's Elsewhere: line is set apart from the bullet above it.

Still for Derek: found in review, already on master, not fixed here

• H1 cannot explain the row St Sir de Mesnil (family-first) → given "de Mesnil". H1's Except says a never-given particle with a word behind it keeps the family (P1's fold), and de is one. The Accepted sentence the row stands under covers only an ambiguous or bare particle. The real reason is that a title-particle (St) opening the run stops P1's fold; decisions/H1 records it and no rule states it. Master's block text had the same mismatch.
• Smith, Dr King Jr reads title "Dr King", given "Smith", suffix "Jr". H3's give-back does not fire, because H1's comma-with-only-titles reading takes the input first. Whether H3's "part carrying the given name" should cover this shape is open, and I have not verified that it is a defect.

Verification

• uv run pytest: 22,154 passed, 123 skipped, 3 xfailed (at 4be2b4a; doc tests rerun at a839d5a).
• uv run mypy and uv run ruff check: clean.
• The design-docs-reviewer agent ran on the migration, on db4a4f7 and on the decision commits; code-reviewer and comment-analyzer reviewed the PR. Every migration finding is fixed above.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.05%. Comparing base (fb0b5a2) to head (37f3a27).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #779   +/-   ##
=======================================
  Coverage   99.05%   99.05%           
=======================================
  Files          47       47           
  Lines        4642     4642           
=======================================
  Hits         4598     4598           
  Misses         44       44           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.