[derek73]

Closes #835.

Doe, Jane Mrs now gives title Mrs, given Jane, family Doe, where 1.4.0 through 2.5.0 gave middle Mrs (all three wheels checked). It already read that way written Doe, Jane Mrs., and so did Doe, John Jr Mr. The FEC's donor file writes FAMILY, GIVEN MIDDLE SALUTATION and often leaves the period off.

The rule: H8, a salutation ending the given part

After a family comma, the surname's place is marked. A bare word ending the given part can then only be a middle name or a title. A salutation is a form of address, not something a person is named after. So a listed salutation there joins H5's trailing title chain, as a period-marked title already does. One consequence: the bare and period-marked spellings read alike in every comma shape and under every name order. That covers the empty head, the nickname head, particle and connective surnames, second commas, suffixes and other titles.

• Not affected: other title kinds (Doe, John King, Doe, Jane Dr), title/suffix duals (Doe, Jane Ms), and the comma-less writing, which stays H5's (Jane Doe Mrs keeps family Mrs).
• Accepted costs (rules/H8):
  • A middle name that is a salutation reads as the title (Doe, Jane Miss).
  • With nothing before the comma, the one given word moves to the family, as the period-marked spelling has done since 2.3.0 (, Jane Mrs).
  • Under a declared family-first order, the comma-less Doe Jane Mrs keeps middle Mrs. So the family-comma invariant in decisions/two-input-invariants no longer holds for trailing salutations: 738 of 738 pairs agreed on master, none now.

Vocabulary

There's a new behavioral title property, after_given = true. A build check allows it on salutation readings only. It compiles into AFTER_GIVEN_TITLES and a new Lexicon.after_given_titles. It has no v1 Constants attribute; the shim intersects it with titles.

88 salutations carry the flag. A middle name may be a given name or a surname, so a salutation borne as either is excluded, with its counts recorded on the word:

• goodman, herr, herren, master, mister, señora, señor, signor, signore, herra, härra, paní, fru, frú, пан
• the Bengali মোঃ/মো (they abbreviate Mohammad), श्री/শ্রী, and बाबू (Babu)
• hr, after the FEC read below

A non-Latin word is weighed by its romanization only where its bearers write that script. The line at roughly 200 US bearers is a judgment call. decisions/H8 has the counts and the reasoning.

Parser

• trailing_titles (and tail_reading) take a salutations flag that admits a word tagged vocab:after-given-title by classify.
• Assign's family-comma walk passes it, and so do the two group-side predictors of that part: the particle chain's bound (trailing_run_start) and the bound given-name join's decline. All three therefore read the part the same way.
• The flag is an inline tag test, so no new call frames: the serial benchmarks didn't move.

Measured (decisions/H8 records each with its recipe)

• Grid: across 9 comma heads × 21 given parts × 7 salutation spellings × 6 tails × 3 name orders, all 23,814 bare rows read exactly as their period-marked twins (fields and ambiguity kinds). No period-marked row moved against master. The reviewer reproduced this on a 128,700-row grid of its own.
• Committed corpora: no name moved. The snapshot gains only the new rule examples, and every CLDR/GND/Wikidata agreement view matches its record.
• fec-individuals: 12,381 of 1,893,964 names move: MR 9,339, MRS 1,908, MISS 1,108, MR.. 24, plus 2 joining a title already there. 385 of them drop a CONJUNCTION_OR_INITIAL report, since the letter no longer has a name word behind it.
• ukec-individuals: no moves.
• Not decided here: in 59 FEC names, a lone I or V is left at the end of the given part and reads as a generational suffix (DOE, JANE I MRS). DOE, JANE I and DOE, JANE I MRS. already read that way on master, under Issues with names in the format Last, First Middle Suffix where Suffix is I #144's rule for a trailing letter after a family comma.

Compatibility

A Lexicon or Parser pickled by an earlier release no longer loads (ValueError naming the missing field), as happened in 2.1 and 2.3. The release log says so.

Review

Three review rounds, each fixed in its own commit:

• Design-docs review, twice. Round one found ten problems: stale "read as the comma-less name reads it" text in H5, P5 and the AGENTS.md Gotcha, the bare-word doctrine stated without the exception in four more places, vetting gaps (बाबू, fru), the unrecorded family-first disagreement, an empty-head row outside H8's When, and small count errors. Round two, on the fix commit, found five, including a false "has always read it" (true only since 2.3.0).
• PR review (code, tests, comments). The code reviewer found no bugs. The test reviewer's mutations showed three behaviors no test caught: the Lexicon field's API, the classify guard, and the flag's hand-off to tail_reading's second chain pass. Each now fails a test when broken (a field/API and off-switch test, an orphan test, and a bare-vs-period-marked parity sweep over 2,112 pairs). Eight comment items were fixed, and a final review of that fix commit tidied five more.

AGENTS.md gains one lesson: vet a non-Latin word by its romanization. The lone I/V this change leaves ending the given part (59 FEC names) is filed as #877.

Checks

• uv run pytest: 24,212 passed. -m serial: 39 passed.
• mypy and ruff are clean, and tools/vocab/build.py --check reports the compiled vocabulary current.
• Sphinx doctest and HTML builds print nothing, and README doctests pass.
• A py3.15 subset passed, but on an earlier commit (before the review fixes).
• master (Phase 4: order and shapes, the rules a pack can choose, and the output model #754) merged in; the snapshot differs from master's only by this PR's 24 rule examples.
• Re-recorded: rules_budget.json (H8 is new, H5 and P5 grew), corpus_rules.jsonl, snapshot.jsonl, and the field-sweep claim count (1,400 on master → 1,403).
• The migrate.rst "Changed in 2.6" entry is left for the 2.6.0 release, with the cycle's others.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.07%. Comparing base (5a75c1d) to head (65f4c65).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #876   +/-   ##
=======================================
  Coverage   99.07%   99.07%           
=======================================
  Files          47       47           
  Lines        4779     4781    +2     
=======================================
+ Hits         4735     4737    +2     
  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.