Peel a title written before a family comma into the title - #706

Closes #701.

What changes

A run of listed title words opening the part before a family comma now goes to title instead of the family. Every release from 1.4.0 through 2.4.2 read it into the family (checked against the 1.4.0, 2.0.0, 2.3.0 and 2.4.2 wheels).

The run is defined in the new rule rules.md#H6:

• Vocabulary only: H2's period shape does not fire in a surname's slot.
• It stops at a word that is a particle too (St, Freiherr, bare or period-marked). That keeps C1's 1919 legal-surname reading.
• It stops where nothing but post-nominals stands behind it (H3's floor).
• It leaves the family its last piece.

Accepted cost: King Smith, John gives title King, as Greve Anna already does under H1.

Docs

• rules.md: adds H6, rewrites the Accepted clauses of H1, H2 and H3 and C1's Freiherr clause, which stated the old reading, and qualifies the statements of C1, S2 and W3. The interacts: links run both ways.
• decisions.md#H6 records the decision, the review finding, and the measurements.
• Also updated: AGENTS.md (the gotcha, plus a sweep lesson about second-position duals), usage.rst (a fifth bound on the inference, with doctests), and a release-log bullet.

Measured (py3.11, against master eb2c1fa5)

• Snapshot: 20 corpus entries move, all one shape (a title leaving the family). 18 of those names are held only by corpus_cldr.jsonl.
• CLDR agreement: 29 fields now agree and 28 more get closer to CLDR's value, with none regressing.
• After merging master (Add owura and bapak to TITLES (#662) #705 listed bapak): one more move of the same shape, Bapak Adinata, Dwi Putro bin MP → title Bapak.
• Frames: the reference name 366 → 366, Smith, John 161 → 161 and Smith Jones, John 203 → 203: a part no title opens pays nothing. A titled part costs +1 (Dr. Smith, John 200 → 201). A /simplify round took out a re-ask of pieces past the floor, so a long post-nominal run costs +1 rather than one frame per member. No reading moved on an 88,740-head grid.

Review

A design-docs review found the first stop too broad: it asked "is the next word suffix vocabulary". A title/suffix dual second in a run blocked the peel and then read as a suffix inside the family (Rev. Sr. Smith, Mary), and so did Mc (Dr. Mc Donald, John). It found this by comparing T Smith, John with T John Smith over every title word. The stop is now H3's question instead. The same review found W3, C1 and S2 prose still calling the pre-comma part wholly family, and one-way interacts: links; both are fixed. A second round (code, test-coverage and comment reviewers) found no parse defects and led to the changes below. A focused review of those fix commits found one false claim about 1.4.0, now corrected.

• gov added to TITLES (your call). Lt. Gov. Smith, John had split into title Lt., family Gov. Smith. The cost: 236 US Census surname bearers, so a surname written first loses it (Gov Sokha). Recorded in decisions.md#H6, with its own release bullet.
• Dr. PhD Smith, John keeps family PhD Smith (your call). 1.4.0 and master read it family Dr. Smith, suffix PhD; the branch reads it as the untitled PhD Smith, John reads. No row, since it's garbage input.
• H6's Accepted clause now names the particle stop as what keeps Freiherr, so Graf Lambsdorff, Otto reads title Graf. Whether the German ranks should all read as freiherr does goes to its own issue.
• New rows: joined title units, Lt. Gov., Dr. King, a FAMILY_FIRST_GIVEN_LAST twin and bare Gov. These cover two mutants that had survived the whole suite.
• Wording fixes in rules.md, AGENTS.md, usage.rst and the release log.

Full suite: 20526 passed. ruff and mypy are clean.

🤖 Generated with Claude Code