Correct AGENTS.md's particles.py bullet to rules.md#P1's one-site fold

[derek73]

AGENTS.md's particles.py bullet (Architecture → Configuration layer) still described the pre-#467 fold, with two sites: the opening piece and the given position. Its example, "Mesnil de" under FAMILY_FIRST folding into the family, is false. rules.md#P1 states "Mesnil de" family-first → given="de", and the code in _post_rules.py agrees: one site, the piece that opens the name.

Docs only. No parser behavior changes, so there's no rules.md amendment.

What the rewrite says. Every example was measured on this tree on 2026-10-04 under all three name_orders:

• One site: the opening piece. A lone never-given particle there, with another name word behind it, starts the family. Titles are walked past: Sir de Mesnil gives family de Mesnil under every order.
• How far the fold reaches depends on the order.
  • Default order: the rest of the name (de Mesnil Jean → family de Mesnil Jean).
  • Either family-first order: the particle run plus one name unit (de Mesnil Jean → family de Mesnil, given Jean).
  • After a family comma: the rest of that part, whatever the order (Smith, de Mesnil Jean → family Smith de Mesnil Jean).
• The leading particle stays a lone piece. group doesn't chain it forward. The fold walks the particles after it, so de la Vega gives family de la Vega under all three orders.
• The given position is not a site, and neither is the inside of the name. Mesnil de under either family-first order gives given de. Juan de la Vega under family-first gives family Juan, given de la Vega, with no ambiguity recorded.
• Ambiguities. van Gogh records PARTICLE_OR_GIVEN under all three orders. de Mesnil and Vincent van Gogh record none.

Other claims in the old bullet that were wrong:

• A bare de "keeps given de". That's only true under the default order. Under both family-first orders it's family de, placed by position.
• Juan de la Vega under FAMILY_FIRST: "1b declines and reports it". There's no given-position check left to decline, and nothing is reported.
• "Reading it as leading-only is how the FAMILY_FIRST bug got in". Since Ménil Christophe de under FAMILY_FIRST drops the tussenvoegsel, and de Mesnil Juan reads Juan as part of the surname #467, leading-only is the rule.
• "Rule 1b" is a retired number. The bullet now names rules.md#P1, which is what decisions.md's numbering map gives for it.

Dropped rather than re-verified: the claim that #359 records the Juan de la Vega case as intended, and the pre-#367 history of Sir de Mesnil. Both are history, and the current behavior is stated instead.

Tests: doc tests (test_doc_*, test_rules_doc*) pass, 771.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.98%. Comparing base (aa401d8) to head (ed201c0).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #605   +/-   ##
=======================================
  Coverage   98.98%   98.98%           
=======================================
  Files          45       45           
  Lines        4238     4238           
=======================================
  Hits         4195     4195           
  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.