[derek73]

Closes #1069

Case repair lowercased every particle that had no case mask. That gave JOHN ST CLAIR → John st Clair and JOSE SAN MARTIN → Jose san Martin, and #1064 extended it to the German ranks it made particles: OTTO GRAF LAMBSDORFF repairs to Otto Graf Lambsdorff under the 2.8.0 wheel but to Otto graf Lambsdorff on master dadf39b.

The change. The saint and rank vocabulary kinds compile into a new internal set, CAPITALIZED_PARTICLES, and case repair (_render._lowers_as_particle) no longer lowercases a word in it. The set is built from the kind rather than from a mask on each word. So a new saint or rank word, whether a vocabulary file's or a caller's Lexicon.add(w, kind="saint"), cannot miss it. The cased masks #725 gave sta and sto for the same purpose are dropped, leaving one mechanism. A recorded mask still outranks the capital (rules/R4's #604 clause).

Measured.

• Committed snapshot (snapshot.py, against dadf39b): 15 entries move, all in the forced case view. 14 now give the input's own spelling: Jean St Clair, Jill St. John, Ste Marie, Diego Fernández de San Pedro, Domingo Santa María González, Freiherr von Berg, Ed, Graf Lambsdorff, Otto, San San Chiou, … The 15th is the accepted cost below.

• Local corpora (snapshot.py --local SOURCE --base <dadf39b7 checkout>), entries that differ:

  corpus	entries	case view
  fec-individuals	896	capitalized
  ctgov-officials	143	139 capitalized_force, 4 capitalized
  tse-donors	323	capitalized

  In FEC the new capital is on St 754 times, San 88, Santa 40, Ste 8 and Graf 9. Examples: Janet st Jean → Janet St Jean, Lane santa Cruz → Lane Santa Cruz. No role moved.

• Against the 2.8.0 wheel (PYTHONSAFEPATH=1, paths asserted): OTTO GRAF LAMBSDORFF, MARION GRÄFIN DÖNHOFF, JOHANNES FÜRST VON THURN UND TAXIS and GEORG FRIEDRICH PRINZ VON PREUSSEN now repair exactly as under 2.8.0. JOHN ST CLAIR repaired to John st Clair on 1.4.0 and on every 2.x wheel through 2.8.1.

Accepted cost. Under a family-first order with force=True, Suzuki-san → Suzuki-San. The romanized honorific had been lowercase only because san is a saint particle. The spaced Suzuki san already repaired to Suzuki San. The vocabulary records no romanized honorific. Keeping the lowercase only where a saint word ends a hyphenated word was considered and declined.

Docs.

• rules/R4: the particle bullet gains the clause, with its examples and an Accepted bullet for Suzuki-san.
• decisions/R4: a Standing bullet and a 2026-11-04 Should case repair write St Clair and San Martín with their capital? JOHN ST CLAIR repairs to John st Clair #1069 entry.
• decisions/H6: Should Juan Sta. Maria read family Sta. Maria? Today it reads middle Sta., family Maria #725's entry gets a head note marking its masks superseded.
• release log bullet, a customize.rst doctest, AGENTS.md, the vocab README kind table, and the re-recorded snapshot.jsonl, corpus_rules.jsonl and rules_budget.json.

Checks.

• Suite: 37,265 passed on 3.11 and on 3.15. The serial benchmarks, mypy, ruff and the Sphinx doctests pass too.
• The new tests sweep every word in the set, and test a caller's word of each kind with a plain particle control. Each fails with the skip removed.
• A code review and a design-docs review ran. The design-docs review's three findings are fixed in the second commit.

🤖 Generated with Claude Code