fix(group): a given-name title licenses the bound-given join (#369)

[derek73]

Closes #369 — the one within-cycle regression from #367 (Sheik Abu Bakar read given Abu, family Bakar on master where 2.1.0 read given Abu Bakar), fixed by the rule that should have produced that reading all along.

The change

P5's bound-given join (abdul, abu, …) needs a name word to spare for the family, and a title is not a name word, so behind one two words never joined. But sheik is a given-name title — the vocabulary H1 reads as "a given name follows", which is what keeps Sir John a given name with no family. Behind a given-name title the join now uses the post-comma LENIENT reserve.

input	before	after
Sheik Abu Bakar	given Abu, family Bakar	given Abu Bakar (restored)
Sheik abdul salam	given abdul, family salam	given abdul salam, family ''
الشيخ عبد الله	given عبد, family الله	given عبد الله
Dr. abdul salam	unchanged	unchanged — a title that addresses by family keeps STRICT
Sir abdul van der Berg	unchanged	unchanged — the licence is for two words; a particle chain is the family
Sheik abdul Jr Smith	unchanged	unchanged — a suffix is not a name word either

Four constraints keep it narrow:

• The title run is keyed through _title_key exactly as post_rules keys H1, so P5 and H1 cannot read one run two ways (mr sir and sir and mrs are not given-name titles to either). Pinned as a parametrized contract over eleven title runs.
• The piece the join takes must be a single word — not a particle chain (found by the design-docs review) and not a suffix piece (found by the PR review; abdul Jr Smith Berg reads given abdul Jr — P5's join takes a suffix as "the word after" the bound word #421's shape, kept away from titled names).
• Only STRICT relaxes; the family comma's DISABLED segment never reaches the block and LENIENT is already the floor.
• absorbs_marker still holds under the lowered reserve.

With the join taking Abu before it can be a lone leading particle, Sheik Abu Bakar no longer reports PARTICLE_OR_GIVEN. That is correct (a bound word read as the bound word is not a fork) and is now stated in P5 as the precedence over P4 where the join fires — Abu Bakar Salim had exercised it silently all along, and bare Abu Bakar keeps P4's reading and its report.

Accepted, not scoped: the join is order-blind, so under FAMILY_FIRST / FAMILY_FIRST_GIVEN_LAST the licensed pair lands in the family with no given name (Sheik abdul salam → family abdul salam), exactly as Sir John reads family John there — H1's exemption never runs under a family-first order, so the two rules still agree. Scoping the licence to the given-first read would need the effective order, which assign settles after group. Recorded in rules.md (family-first rows), decisions.md and the release bullet; two corpus names move under each family-first order, measured by hand (the gate reads the default order).

Verification

• Suite: 5199 passed, 212 skipped, 10 xfailed (the Sheik abdul salam splits the bound given name that abdul salam ahmed joins #369 xfail goes live); ruff and mypy clean. Every commit is green on its own.
• Differential gate exits 0 at 1.4.0, 2.0.0 and 2.1.0. Two corpus names move in fields at every baseline — Sheik abdul salam and its Jr spelling from the rules examples; Sheik Abu Bakar is byte-identical to every baseline and differs from 2.0.0/2.1.0 in its report alone. The three old fix(#369) ledger rules classified the regression; they are rewritten to classify the fix (two narrow rules at 2.x — no Latin alternation needed), and the 1.4.0 one is what keeps Sheik abdul salam off the fields-only suffix-routing catch-all.
• Reviews run on the branch: design-docs-reviewer (four findings), then code-reviewer, pr-test-analyzer and comment-analyzer (three important + a handful of test/comment items). All addressed in commits 3–5; nothing deferred except the abdul Smith V loses the family name — the bound-given join counts an initial-shaped suffix as a name word to spare #401 note below.

Commits

1. fix(group) — the licence, tests (group + parser), ledgers, release log, AGENTS.md.
2. docs(design) — rules.md#P5/#H1/#P4, decisions.md#P5, regenerated corpus_rules.jsonl, reach pins.
3. fix(group) — the licence declines a suffix piece; comment corrections from review (the "no corpus name has the shape" claim was false on this very branch: rules.md examples are corpus names, and the gate witnesses the single-word clause).
4. test(parser) — the H1/P5 agreement contract, the ambiguity-report assertions, the Arabic-script row.
5. docs(design) — the family-first Accepted outcome, the suffix/P4 statements and examples, the abdul Smith V loses the family name — the bound-given join counts an initial-shaped suffix as a name word to spare #401 note, ledger wording, reach re-pins.

For the reviewer

• abdul Smith V loses the family name — the bound-given join counts an initial-shaped suffix as a name word to spare #401 widening, not fixed here: Sir abdul V now reads given abdul V — the reserve's initial veto counts a bare V as a name word, and the lowered threshold lets that count fire where STRICT did not. Recorded in the decisions entry; worth adding the shape to abdul Smith V loses the family name — the bound-given join counts an initial-shaped suffix as a name word to spare #401's verification when that fix lands.
• The 2.2.0 release-log preamble's "only one of them reaches the default name order" was already stale before this PR (the Ursula von der Leyen geb. Albrecht loses the maiden name — a particle chain swallows the marker where a suffix stops it #399/Abd van der Berg née Jones reports no family name — the bound-given join counts words the maiden name takes away #411/juan y garcia nee jones loses the family name — P3's word count includes the words the maiden name takes away #418 fixes are default-order too) — release-prep prose, left alone.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.56%. Comparing base (58ab662) to head (ca64ed0).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #422   +/-   ##
=======================================
  Coverage   98.56%   98.56%           
=======================================
  Files          44       44           
  Lines        2993     2997    +4     
=======================================
+ Hits         2950     2954    +4     
  Misses         43       43           

☔ 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.