Count a lone particle with the run behind it as a name word (#627)

[derek73]

Summary

#602's suffix run starts at a credential only with two name words in front of it, and "a lone particle is not a name word for the count" (rules.md#S2), because in de Mesnil the particle and the word behind it are one surname. Since #620 the count is taken in the particle chain's units, so a particle with something joined to it already counted. That left the rule firing in exactly one place: a bare particle with the credential right behind it, where nothing joins it.

Input	Before	Now
John von und zu PhD Jones	family 'von und zu', suffix 'PhD Jones'	same
John van der PhD Jones	family 'van der', suffix 'PhD Jones'	same
Smith, John von PhD Jones	family 'von Smith', suffix 'PhD Jones'	same
John von PhD Jones	middle 'von PhD', family 'Jones'	family 'von', suffix 'PhD Jones'

Decided by Derek: the particle stops at the known suffix. The input is malformed, so the reading is whichever one the shared count already gives the other forms, not a special case. The existing clause in run_start gains its own limit, in the one loop both callers share (with the chain's units and without): a lone particle counts as a name word of its own when the run starts right behind it. A particle followed by a name word still counts once with that word (de Mesnil, and the existing row van la Smith Secretary Jr. Dr. Smith).

Declined: deleting the clause outright, so that every unit counts. It also counted de Mesnil <credentials> as two words, and it split the two callers: a leading van counted before another particle but not before a name word. The case row above caught it.

Measured (decisions.md#S2, 2026-10-10)

Against master 195c408, py3.11, nameparser.__file__ asserted on each side:

• S2 stress grid (the Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614 entry's recipe, 324,552 parses): 2,432 role moves and no report-only move. Every move is a lone particle (in that grid van, or a dual do/mc/vd in any case) with an unambiguous credential right behind it becoming the family, and the run taking the rest. The grid's comma shapes don't move.
• Beyond the grid: the review found the same move, as intended, for other particles (von, de, du, la, bin) and for a head before a suffix comma (John von PhD Jones, Jr. → family 'von', suffix 'PhD Jones, Jr.').
• Fingerprint of every corpus name and case text at 195c408 under the three orders: 0 moves. This PR adds one text, its own example, which moves.
• 1.4.0 and 2.3.0 wheels: both read middle 'von PhD', as they read von und zu PhD. The joined forms changed this cycle with Should a name word after a credential join the suffix? John Smith PhD Jones reads middle Smith PhD #602.
• Five differential gates: all exit 0. The new rules.md example is explained by a fix(#627) rule in each ledger.
• Frames: plain names are unchanged (Smith, John 161, Dr. Juan de la Vega III 304 on py3.11), and the suite's frame guards pass unchanged. The moved shapes cost more: John von PhD Jones 242 → 266, the run's own reading.
• The caller without units (trailing_start) shares the expression, so the two callers can't disagree. No reading shows that half, but it isn't free: for Smith, John von PhD Jones it returns 2 where master returned 4, so group's chain takes its stopped path and asks trailing_start again, reaching the same reading at 16 more frames (339 → 355). It's a rare garbage shape no guard sees, so the shared expression wins over a per-caller copy (AGENTS.md's frames rule).

Changes

• nameparser/_pipeline/_pieces.py: run_start's lone-particle clause and docstring.
• rules.md#S2: one clause (the particle "counts as a name word of its own", true in every order), plus the example pair John von PhD Jones → family 'von', suffix 'PhD Jones'.
• decisions.md#S2: new 2026-10-10 entry (decided, declined, measured). The Should a name word after a credential join the suffix? John Smith PhD Jones reads middle Smith PhD #602 entry points at it.
• tests/v2/cases.py:
  • Row a_lone_particle_before_the_run_is_the_surname (fix(#627)). The field sweep's recorded claim count moves from 1155 to 1158, the row's three orders.
  • Row a_particle_run_before_the_run_stays_one_name_word (Freiherr von vd PhD Jones, parity): the flag is asked only of a piece that opens a unit. A variant asking it of every piece passed the whole suite until this row.
• tools/differential/: the regenerated corpus_rules.jsonl and a fix(#627) rule in all five ledgers. test_ledger_guards.py records the new claims, plus fix(initials-per-word)'s reach growing by the new name (123 → 124).
• No release-log bullet: the run reading is new in 2.4 (Should a name word after a credential join the suffix? John Smith PhD Jones reads middle Smith PhD #602), and the shape is malformed input.

Review

• 3464370: the change.
• 8ea6865: fixes the first review: the guard row, "counts as a name word of its own" for the order-specific "is the surname", and the decisions entry's grid limits, units-less half and fingerprint population. abdul von PhD Jones and Sir abdul von PhD Jones move consistently with the rule and are left without rows (malformed input).
• e4b653a: fixes the fix commit's review: the five ledger comments quoted the old rules.md wording ("then it is the surname"; the excerpt check scans only .py, filed as a follow-up task), the frames claims above, and the remaining "is the surname" phrasing in decisions.md.

Closes #627

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.01%. Comparing base (195c408) to head (e4b653a).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #631   +/-   ##
=======================================
  Coverage   99.01%   99.01%           
=======================================
  Files          46       46           
  Lines        4486     4488    +2     
=======================================
+ Hits         4442     4444    +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.