[derek73]

Ships nameparser 2.7.0, vocabulary phase 2 (#752). It adds the title phrase (rules/H9, a title that is also a given name is a title only inside a title phrase), credentials written joined, split, in capitals or in brackets, given names written short, and Ambiguity.alternative. Facade DeprecationWarnings wait for 2.8, as decided.

Release checklist, steps 0–3

• Step 1. snapshot.py --since v2.6.0: 420 entries differ, 2 of them held only by corpus_cldr.jsonl. 304 are names new to the corpus and 15 were removed, which leaves 101 moved readings; 57 of those move only an ambiguity. Every move traces to a release-log bullet (H9 and phase 2j, S6 capitals, S13 brackets, Should John Smith, MD - PhD split its suffixes without a declared delimiter, and can suffix_delimiter be deprecated? #889, Should John Smith VI MD read like John Smith III MD? #784 numerals, Should John Smith Ph. D. Jones read like John Smith PhD Jones, or is the split spelling's reading the better one? #827, S11 agnomens). The v2.6.0 snapshot carries the case views (Has case repair now earned a column in the snapshot, as initials() did? #826), so they are compared there and need no separate compare.py --baseline 2.6.0 run.
• Release log
  • Merged bullets. The eight phase-2j given-name-title bullets are now one. The Shaikh bullet still said "Sheikh and Shekh are still titles", which Read the name Sheikh and its spellings as names outside a title phrase (#752 phase 2j, #853) #912 had reversed.
  • Order. The title phrase leads, and the entry opens with a summary and highlights.
  • Checked against the 2.6.0 wheel. I ran every example in every bullet (about 120 names) on the 2.6.0 wheel and on the tree. All of them hold.
  • Pickle bullet. I pickled every public type under 2.6.0 and loaded it in the tree. Lexicon, Parser and locale packs raise, Policy, ParsedName, Ambiguity and HumanName load. The bullet had missed locale packs; it now names them. The new fields, from a dataclasses.fields diff against v2.6.0: Lexicon gains agnomens, given_name_abbreviations, suffix_words_ambiguous, title_modifiers, title_phrases, title_styles and titles_ambiguous. Ambiguity.alternative carries ADDED_IN, so those pickles still load.
• migrate.rst "Changed in 2.3 through 2.7"
  • New entries. Eleven 2.7 entries, each with a doctest, every "1.4 read X" checked against the 1.4.0 wheel. They cover the phase-2j words (King, Prince, Princess, Sultan, Sultana, Sheikh and its spellings, Mahdi, Bard, Buddha, Siddha, Gyani, Wing, Grand), title phrases, Chas./Ma., numerals VI to X, MD/PhD and Pharm D, Ph. D. Jones, MA Jones, Filho and fils, (JD), and the existing / and Bon/Herren entries, which moved to the end.
  • Not listed: Dean Koontz and Major Harris. They read the same through HumanName on 1.4.0, 2.6.0 and 2.7; only the 2.0 API's new title-or-name report changed.
  • Prince and the v1 knob. The 2.3 Prince entry gets a 2.7 qualifier. A note explains that Constants().titles cannot restore a moved title: the word is still listed there, so adding it is a no-op. The 1.4 dean recipe still works.
  • Pickle break added to the list of 2.x breaks.
• User docs audit (docs/AGENTS.md style)
  • customize.rst. A new "Titles that are also given names" section covers H9, the title-or-name report and its alternative. The marker-field paragraph is now a table, which fixes a claim that given_name_titles, title_modifiers and title_styles sit over titles only. The post-nominal separator section is split by task, with Pharm D doctested. A wrong claim is fixed: declaring " - " does not drop a glued hyphen.
  • usage.rst and concepts.rst. Both gain the title phrase.
• Code-side prose
  • Reads: lines. Checked mechanically against each stage's lexicon./policy. accesses. _extract (suffix_words_ambiguous, Gate generation readings by kind: fils, the Brazilian agnomens, default = false (#752, phase 2h) #890) and _assign (given_name_titles, Carry the reading a report did not take, as Ambiguity.alternative (#752, phase 2m) #902/Name an alternative's field through one slot and one name-piece helper (#752) #905) had outgrown theirs.
  • Docstrings. TITLE_STYLES and Lexicon.title_styles claimed styles are titles only, but seven sit in titles_ambiguous and are consulted there. The Lexicon docstring now names title_phrases as a third multi-word-exempt field.
  • Smaller fixes. The AmbiguityKind.TITLE_OR_NAME wording and a cases.py note.
  • Messages. No raise or warn text added this cycle hands the reader code to paste.
• Design docs and AGENTS.md (read, not grepped)
  • VI-to-X examples. rules/P5's Accepted clause, AGENTS.md's tail_reading gotcha and decisions/H5's standing line cited examples that Should John Smith VI MD read like John Smith III MD? #784 made read as suffixes. They now show the V shape that remains; the dated log entries are kept as history.
  • C2 and S8. Both now name H9's exception (John Smith, Jr., Dean keeps suffix Jr., Dean).
  • Open lines and AGENTS.md. generational-words' Open line drops Should José Silva Júnior read Júnior as a suffix, as José Silva Junior does, and where do Filho, Neto and Sobrinho belong? #843, closed and decided in decisions/S11. AGENTS.md's GIVEN_NAME_TITLES "subset of TITLES" claim and its honorific caution are fixed.
  • rules_budget.json re-recorded for P5 (shrank) and C2/S8 (grew), all from these edits.
  • Left alone, pre-existing. The P background calls Mc Donald and Ste Marie read the particle as the given name #360 open (it closed 2026-08-18), and S11's 2026-10-25 "NOT DONE" line lacks a superseded marker.
• README.rst leads with 2.7. python -m doctest README.rst passes and readme_renderer renders it.
• Dependabot: no open PRs, no open security alerts.
• Milestone 2.7 (Update LICENSE to be a bit more explicit #47): 52 closed, 0 open. No issue or PR closed since 2.6.0's publish is missing from it.
• Steps 2–3. PRE_RELEASE = "", CITATION.cff 2.7.0, release log stamped October 27, 2026. Restamp it if the tag lands on a later day.
• AGENTS.md checklist. It gains the pickle-break check that found the locale-pack gap.

Review

• First round, design docs. The new C2 and S8 clauses dropped H9's phrase limit: past the second comma, Dean of Students is still a title, while a bare Dean stays a suffix. AGENTS.md still claimed GIVEN_NAME_TITLES ⊆ TITLES at two more sites. Both fixed in 9e1b297.
• First round, claims. I checked about 90 names against the 1.4.0 wheel, the 2.6.0 wheel and the tree, and found no false history claim. One defect: the merged phase-2j bullet let "a bare one reverses after a comma" cover Vice and Grand, which stay titles there, and it had dropped the Wing. clause. Smaller fixes: README said King "keeps" a first name, and the first highlight had two before/after pairs. All fixed in e27aff9, and AGENTS.md's release-log guidance now says to carry a merged bullet's limits over.
• Second round, on the fix commits. A double comma left by the highlight trim, fixed. The rest of the fix commits checked clean, including the comma reversal for every listed word and the phrase/bare split in the C2 and S8 slots.
• Derek's edits. The 2.7.0 section's summary now introduces 2.7 as the first release towards the vocabulary model RFC, and README's announcement follows it. The section names the fields given and family rather than first and last. The title-phrase highlight now shows the contrast case, Justice Smith giving given Justice.

Local checks:

• full suite on 3.11 and on 3.15: 32,127 passed each
• -n0 -m serial: 43 passed
• mypy and ruff
• build.py --check
• Sphinx doctest: 477 tests, 0 failures
• Sphinx HTML with -W

After merge: tag v2.7.0 on the squash commit, gh release create, approve the pypi environment, close milestone 2.7. Then a PR for steps 8–9: DEFAULT_BASELINE 2.7.0, VERSION 2.8.0 with PRE_RELEASE = "dev". No "Next Release" milestone unless Derek asks for one.

🤖 Generated with Claude Code